Compare commits

..

6 Commits

Author SHA1 Message Date
Ben Stull b3f1b15f65 Release 0.12.0: CloudFlare Turnstile on OTC email-entry 2026-05-28 03:52:45 -07:00
Ben Stull 6fb68a95c7 Release 0.11.0: trust device for 30 days 2026-05-28 03:46:24 -07:00
Ben Stull 7872b921ed Release 0.9.0: admin user-management page + new-request notifications 2026-05-28 03:39:25 -07:00
Ben Stull de28272914 Release 0.14.0: public /docs user guide (DOCS.md served at /docs)
Ships DOCS.md and the /docs route — a plain-prose translation of
SPEC.md for users, distinct from the binding spec. Originating need
was the admin-vs-owner role distinction on /admin/users (§6.1);
response is a single guide that covers the framework's user-facing
surfaces end-to-end (roles, proposing, branches, PRs, graduation,
notifications). Mirrors /philosophy: markdown file at repo root +
sibling backend loader + MarkdownPreview render. No schema migration.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-28 03:08:28 -07:00
Ben Stull 55beba5c0a Release 0.10.0: user-set passcodes (OTC stays as fallback)
After a successful OTC sign-in, a contributor can set a passcode
(4-20 characters, bcrypt-hashed) and use email + passcode for
subsequent sign-ins. OTC remains the structural fallback: five
consecutive failed verifies lock the passcode path for 15 minutes
(HTTP 423), and a forgotten passcode is recovered by requesting a
fresh code. Migration 015_passcode.sql adds four nullable columns
to the users table; existing rows pass through as OTC-only and can
opt into a passcode from a new Sign-in tab in /settings/notifications.
The /login surface is extended to a five-step flow (email → either
passcode or OTC code → optional post-OTC passcode offer → optional
set-passcode). SPEC corrections per §19.3 rule 2: §6 names the
three auth paths, §14.1 documents the stepped login flow, §17 lists
the four new /auth/passcode/* endpoints, §19.2 surfaces four new
candidates and refreshes the cross-refs.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-28 02:39:27 -07:00
Ben Stull ca8ba69acb Release 0.8.0: open beta-access request flow (first/last/why)
Replaces the v0.3.0 / v0.7.0 allowed_emails admission gate with an
admin-grant flow (roadmap item #6, SPEC §6.1 / §6.2 / §14.1 / §17).
Any valid email can sign in via OTC; a fresh user lands in
permission_state='pending' with a captured first/last/why profile,
and an admin grant flips them to 'granted' before write endpoints
accept them. Grandfathered users pass through the migration with
the column default 'granted' so existing contributors are unaffected.
The allowed_emails table stays in the schema as a fast-path bypass
pending v0.9.0's admin user-management page (item #7).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-28 02:25:59 -07:00
34 changed files with 5881 additions and 385 deletions
+810 -95
View File
@@ -23,6 +23,438 @@ skip versions are the composition of each intervening adjacent
release's steps in order — no A-to-B path is pre-computed beyond
that.
## 0.14.0 — 2026-05-28
**Minor — no operator action required; new optional env var.** This
release ships `DOCS.md` and the `/docs` route — a public-facing user
guide that translates `SPEC.md` into plain prose for readers,
proposers, and contributors. The originating need was the
admin-vs-owner distinction on the `/admin/users` surface (the §6.1
role separation was load-bearing but only documented in spec voice);
the response was a single guide that covers the framework's user-
facing surfaces end-to-end. Mirrors `/philosophy` end-to-end: a
markdown file checked into the repo root, served by a sibling backend
loader, rendered with `MarkdownPreview`. No schema migration. No
required env-var changes. The new "Docs" header link sits alongside
the persistent "About" link from §14.3 and is reachable by anonymous
viewers per the same v0.3.0 anonymous-read contract.
### Added
- **`DOCS.md`** at the repo root — the user-facing guide. Covers
reading anonymously, signing in, proposing an RFC, super-drafts vs
active RFCs, the discussion-vs-contribution distinction (§10.10),
working on a branch (contribute mode, AI proposals, manual edits,
flags, branch visibility, contribute grants, hygiene), opening and
reviewing PRs, graduation (§13), withdrawal and reopening, the AI
participant (§6.6 / §6.7 / §18), notifications and watch states
(§15), and the full roles-and-permissions story (§6 in plain
prose: anonymous / contributor / admin / owner, per-RFC
owners + arbiters, per-branch contribute grants, the write-mute,
and the three structurally distinct "mutes"). Framework-neutral —
no deployment-specific names or corpus references; consistent with
`CLAUDE.md`'s separation-of-concerns rule.
- **`backend/app/docs.py`** — sibling loader for `philosophy.py`.
Reads `DOCS.md` from the repo root with the same disk-first,
in-process-cached, `refresh()`-on-demand shape. Optional
`DOCS_PATH` env var points at an alternative source (e.g. a
meta-repo working-tree clone) for deployments that prefer that.
- **`§17` endpoint** — `GET /api/docs` returns
`{ "body": "<DOCS.md verbatim>" }`. Anonymous-reachable, same
contract as `GET /api/philosophy`.
- **`frontend/src/components/Docs.jsx`** — the `/docs` reading
surface. Mirrors `Philosophy.jsx`: chrome with Back / "USER GUIDE" /
Home affordances, body rendered through `MarkdownPreview`.
### Changed
- **`backend/app/api.py`** — imports `docs as docs_mod` alongside
`philosophy` in the relative-import block; registers the new
`GET /api/docs` handler immediately after `GET /api/philosophy`.
- **`frontend/src/api.js`** — exports `getDocs()` alongside
`getPhilosophy()`. Same fetch shape, different endpoint path.
- **`frontend/src/App.jsx`** — imports `Docs` alongside `Philosophy`,
registers the `/docs` route alongside `/philosophy`, adds the
persistent "Docs" header link alongside "About", and adds the
`DocsWithSidebar` chrome wrapper alongside `PhilosophyWithSidebar`.
### Upgrade steps (from 0.13.0)
- You **MUST** rebuild the frontend and restart the backend after
upgrading so the new `/docs` route, the new endpoint, and the new
loader are picked up. `frontend/package.json#version` and `VERSION`
both move to `0.14.0`. No schema migration; the new endpoint
serves a checked-in file.
- You **MAY** set `DOCS_PATH` to an absolute path if your deployment
hosts `DOCS.md` outside the framework's repo (e.g. as a sync target
from a content repo). Unset is supported — the framework's
`DOCS.md` at the repo root is the default, mirroring how
`PHILOSOPHY_PATH` works for `/api/philosophy`.
- You **MAY** customize `DOCS.md` for your deployment if you want
deployment-specific phrasing layered on top of the framework's
guide. The file is a regular markdown source; standard `vim`/`git`
edits suffice. Framework upgrades that ship a new `DOCS.md` will
show as a normal merge in your deployment-overlay layer.
## 0.13.0 — 2026-05-28
**Minor — schema migration required; new optional env vars.** This
release ships the cookie / privacy consent surface (roadmap item #11,
SPEC §14.5 / §14.6). Every viewer — authenticated and anonymous alike —
now sees a non-modal bottom-of-page banner on first visit asking which
categories of cookies they allow (essential / essential + analytics /
essential + analytics + other). The choice persists in `localStorage`
for anonymous viewers and in a new `cookie_consent` table for
authenticated viewers, with server-side overriding local on sign-in.
The framework also ships default `/privacy` and `/cookies` policy pages
that deployments can layer their own policy URL on top of via two new
optional env vars. No analytics SDK ships in this release — the
consent infrastructure is wired so item #13 (v0.15.0) can read from
`frontend/src/lib/consent.js` when the SDK lands.
### Added
- **Cookie consent banner** (`frontend/src/components/CookieConsentBanner.jsx`).
Non-modal, bottom of viewport. Three single-select choices with
inline descriptions. Visible until the user makes a choice; hides
thereafter. Reachable for revision via the settings surface.
- **Consent helper** (`frontend/src/lib/consent.js`). Exports
`getConsent()`, `hasChosen()`, `onConsentChange(cb)`, `setConsent()`,
`hydrateFromServer()`, `clearLocal()`. Cross-tab sync via the
`storage` event. Item #13's analytics SDK reads consent here before
importing.
- **Privacy and cookies policy pages**
(`frontend/src/pages/Privacy.jsx`, `frontend/src/pages/Cookies.jsx`).
Default minimal policies that describe the framework's stance and
list the cookies the framework sets. Deployments override via the
two new env vars below; the framework's stub always renders above
the link so the framework-level contract stays visible.
- **"Privacy & cookies" tab** in `/settings/notifications` showing
the current consent choice, the recorded-at stamp, and a "Change"
button that re-opens the banner via a custom DOM event.
- **`§17` endpoints** —
- `GET /api/users/me/cookie-consent` — read the current consent
record.
- `PUT /api/users/me/cookie-consent` — write a new consent record.
Upserts a single row per user, stamps `recorded_at` to now,
accepts `essential` for symmetry but always persists it as true.
- **Schema migration** `013_cookie_consent.sql` — new
`cookie_consent` table keyed by `user_id`, three flags
(`essential`, `analytics`, `other_cookies`), and `recorded_at`.
(Renumbered from `012_*` during driver integration because v0.7.0
also added a `012_otc.sql` migration that landed in the integration
order before this one.)
- **SPEC `§14.5` Cookie / privacy consent** — settles the banner
shape, the three-category single-select, the storage shape (local
for anon, server row for authenticated), the precedence rule on
sign-in, and the `consent.js` helper surface for downstream
callers including item #13.
- **SPEC `§14.6` Privacy and cookies policy pages** — settles the
`/privacy` and `/cookies` routes, the framework's stub content, and
the `VITE_PRIVACY_POLICY_URL` / `VITE_COOKIES_POLICY_URL` override
shape.
- **SPEC `§5`** — names the `cookie_consent` table in the canonical
app-tables list.
- **SPEC `§17`** — lists the two new cookie-consent endpoints.
- **SPEC `§19.2`** — surfaces four candidates: policy content via
content-repo file vs env var, GPC / DNT headers, multi-language
consent text, and the item #13 analytics-SDK gating dependency.
### Changed
- **`frontend/.env.example`** — documents the two new optional env
vars `VITE_PRIVACY_POLICY_URL` and `VITE_COOKIES_POLICY_URL`. Unset
is supported; defaults render the framework's stub.
- **`backend/app/api_notifications.py`** — module docstring grew two
endpoint lines; the new endpoints sit alongside the existing
`/api/users/me/*` neighbors.
- **`frontend/src/App.jsx`** — registers `/privacy` and `/cookies`
routes (anonymous-reachable), wires `<CookieConsentBanner>` into
the global chrome, and listens for a `rfc-app:cookie-consent-reopen`
custom event to re-open the banner from the settings surface.
### Upgrade steps (from 0.7.0)
- You **MUST** rebuild the frontend and restart the backend after
upgrading. `frontend/package.json#version` and `VERSION` both move
to `0.13.0` and the build embeds the new env-var contract.
- You **MUST** apply schema migration `013_cookie_consent.sql`. The
migration creates a single new table keyed by `user_id` with three
flag columns and a `recorded_at` stamp. The framework runs
migrations automatically at process start; no manual step is
required beyond restarting the backend so the migration runner
picks the file up.
- You **MAY** set `VITE_PRIVACY_POLICY_URL` to an http(s) URL that
points at your deployment's full privacy policy. The framework's
`/privacy` page renders its built-in stub above a link to the
configured URL. Unset is supported — the stub is sufficient for a
default-config deployment.
- You **MAY** set `VITE_COOKIES_POLICY_URL` to an http(s) URL that
points at your deployment's full cookies policy. Same shape as the
privacy URL.
- You **MAY** announce the new consent banner to your users. Existing
authenticated users will see the banner on their next visit
(because their `cookie_consent` row does not yet exist); their
current sessions remain valid.
## 0.12.0 — 2026-05-28
**Minor — operator action required (new secret + new overlay).**
CloudFlare Turnstile gates the email-entry step of the OTC sign-in
flow against automated abuse (roadmap item #10, SPEC §6.2 / §19.2-
settled). Since v0.7.0 made `/auth/otc/request` the primary human-
auth path and v0.8.0 opened the request endpoint to any valid email,
the OTC dispatch became the natural target for distributed scrapers
fanning out to harvest "this email is admitted vs. this email is
not" timing/bounce signals. The per-email cooldown stops the trivial
back-to-back loop; the Turnstile challenge stops the distributed one
by costing the attacker a browser-side proof-of-humanness on every
request. The challenge runs before the bcrypt hash + SMTP send so a
failed verify spends no rate budget and produces no envelope.
Scope: the widget renders on the email-entry step of `/login` only.
The OTC verify step (where the user pastes the six-digit code) is
already bottlenecked on email delivery and protected by the
five-minute TTL + single-use consume on the row; a second challenge
there would double the rate budget against the same abuse path
without measurably more protection. If bots adapt to defeat the
email-entry challenge specifically — pushing the abuse vector onto
the verify step — a future release adds the second widget. The
widget also renders on the passcode step's "Use a code instead"
fallback dispatch since that route also calls `/auth/otc/request`.
Default policy: `TURNSTILE_REQUIRED=false`. The gate stays open when
the secret is absent — the dev / test path, and the pre-rollout
path while the operator is wiring the secret. Once the secret is in
GCP Secret Manager and the site key is in the overlay, the operator
**MAY** flip `TURNSTILE_REQUIRED=true` so a future config drift on
the secret fails loudly (HTTP 500 "auth misconfigured") instead of
silently disabling abuse defense.
No schema migration — Turnstile siteverify is stateless.
### Added
- **`backend/app/turnstile.py`** — the siteverify caller. POSTs
`secret` + `response` (+ optional `remoteip`) to
`https://challenges.cloudflare.com/turnstile/v0/siteverify` and
returns a `VerifyOutcome` (`ok` boolean + `reason` enum:
`ok` / `skipped` / `misconfigured` / `missing-token` / `failed` /
`network`). Tunables read from env at call time so tests
monkeypatch cleanly: `CLOUDFLARE_TURNSTILE_SECRET`,
`TURNSTILE_REQUIRED`, and (test-only) `TURNSTILE_SITEVERIFY_URL`.
- **`frontend/src/components/TurnstileWidget.jsx`** — the React
wrapper around the official CloudFlare Turnstile JS API. Reads the
site key from `import.meta.env.VITE_TURNSTILE_SITE_KEY`; renders
nothing when the var is unset (the form still submits and the
backend's `TURNSTILE_REQUIRED` policy decides admission). Loads
the CloudFlare script once per page on first widget mount. Cleans
up the widget instance on unmount via `turnstile.remove()` so a
remount produces a fresh challenge rather than reusing a stale,
already-consumed token.
- **Backend tests** (`backend/tests/test_turnstile_vertical.py`) —
five vertical scenarios: happy path (secret + valid token →
admit), siteverify rejects → 400 + no envelope, missing-token →
400 + no envelope, missing-secret-soft (default) → admit, and
missing-secret-hard (`TURNSTILE_REQUIRED=true`) → 500
"misconfigured". All five mock the siteverify HTTP call via
`monkeypatch.setattr(turnstile.httpx, "post", …)`; no real
CloudFlare keys are ever embedded.
- **SPEC `§6.2`** — names the Turnstile gate on the OTC dispatch as
the v0.12.0 settled shape; the §19.2 candidate from v0.7.0 closes.
### Changed
- **`backend/app/main.py`** — `OtcRequestBody` grows an optional
`turnstile_token` field. The `/auth/otc/request` handler calls
`turnstile.verify_token` first, before `otc.request_code`, so a
failed challenge spends no rate budget and produces no envelope.
The handler maps `misconfigured` → HTTP 500, all other failures
(`missing-token`, `failed`, `network`) → uniform HTTP 400 so the
response does not enumerate which leg of the challenge broke.
- **`frontend/src/api.js`** — `requestOtc` accepts a second arg
`{ turnstileToken }` and threads it into the request body. The
positional signature stays backwards-compatible so calls that pass
only an email still type-check.
- **`frontend/src/components/Login.jsx`** — the email step and the
passcode step both render `<TurnstileWidget>`. The submit button
on the email step is disabled until the widget produces a token
(when the widget is enabled at build time); the "Use a code
instead" link on the passcode step has the same gate. A 400 from
`/auth/otc/request` clears the token and surfaces a "couldn't
verify you're human, please retry" status. The fallback-from-
passcode path bounces back to the email step on 400 so the user
gets a fresh challenge in the natural place.
- **`backend/.env.example`** — documents `CLOUDFLARE_TURNSTILE_SECRET`
and `TURNSTILE_REQUIRED` alongside the existing OTC tunables.
- **`frontend/.env.example`** — documents `VITE_TURNSTILE_SITE_KEY`
with the operator wire-up procedure (dash.cloudflare.com →
Turnstile → Add site).
### Upgrade steps (from 0.10.0)
The operator **MUST** create a CloudFlare Turnstile site
(dash.cloudflare.com → Turnstile → Add site, choose "Managed" widget
mode), obtain the site key (public) and secret key (private), and:
- You **MUST** `flotilla secret set ohm-rfc-app CLOUDFLARE_TURNSTILE_SECRET`
(paste the secret key when prompted) before the v0.12.0 deploy.
The framework reads the secret at request time; deploying v0.12.0
without the secret leaves the gate in its default soft-fail state
(every request admitted regardless of token), which means abuse
defense is silently off.
- You **MUST** `flotilla overlay set ohm-rfc-app VITE_TURNSTILE_SITE_KEY <site-key>`
so the frontend build embeds the site key and the widget renders
on `/login`. The site key is public — it travels in the bundle and
appears in every browser — so this is the overlay (non-secret)
layer per the §3 invariant 1 split. Skipping this step leaves
`/login` with no widget; even after the operator sets the secret,
the backend would refuse every request as `missing-token` once
`TURNSTILE_REQUIRED=true` flips.
- You **MUST** rebuild the frontend and restart the backend after
upgrading. `frontend/package.json#version` and `VERSION` both move
to `0.12.0`. No schema migration; Turnstile siteverify is
stateless. The site-key embed is build-time, so the rebuild after
the `flotilla overlay set` is what actually wires the widget into
the bundle the deploy serves.
- You **MAY** `flotilla overlay set ohm-rfc-app TURNSTILE_REQUIRED true`
once you've confirmed a real sign-in works end-to-end with the
widget. The default (`false`) keeps the gate in soft-fail mode so
a missing-secret regression admits requests rather than 500ing
every sign-in attempt; flipping to `true` makes a future config
drift on the secret fail loudly with HTTP 500 instead of silently
disabling abuse defense. The framework's tested path is the
flipped-to-true production shape; the default `false` exists for
the dev / pre-rollout window only.
- You **MAY** customize the Turnstile widget mode (Managed /
Non-interactive / Invisible) from the dashboard at any time
without redeploying — the site key stays the same, and the widget
picks up the mode change on the next page load. The framework's
tested path is "Managed" because it gives the operator a visible
challenge surface to debug against.
If either of the two **MUST** secret/overlay steps is skipped, the
deploy still boots and `/login` still serves; the failure mode is
that abuse defense is off (default `TURNSTILE_REQUIRED=false`) or
every sign-in attempt 500s (`TURNSTILE_REQUIRED=true` flipped while
the secret is unset). The driver pauses the wave at the secret/
overlay gesture so the operator confirms both are in place before
the framework version pin moves.
## 0.11.0 — 2026-05-28
**Minor — schema migration required; no new env vars.** This release
ships the "trust this device for 30 days" gesture (roadmap item #9,
SPEC §6.2). After a successful OTC or passcode sign-in, the user
can check a single checkbox to mint a server-issued opaque
device-trust token; the token rides as a long-lived HttpOnly +
Secure + SameSite=Lax cookie, and the matching row's hash lives in a
new `device_trust` table. On a subsequent visit, the cookie is
presented at `POST /auth/device-trust/start` — if a non-expired,
non-revoked row matches, the session is re-established without
another OTC / passcode roundtrip. A new `/settings/notifications`
"Trusted devices" section lists active rows (created-at, last-seen,
expiry, rough UA label) with per-row "Revoke" and a "Revoke all
devices" button. The cookie is "essential" per the v0.13.0 cookie-
consent contract — it is part of authentication, not analytics — and
is set regardless of the user's analytics / other-cookies choice.
The session model gains a cookie, not a session-store change: the
existing `rfc_session` cookie still carries the in-flight session
state; the new `rfc_device_trust` cookie is consulted only by
`/auth/device-trust/start` to bootstrap a fresh session on a return
visit. The raw token only ever lives in the outbound `Set-Cookie`
header and the inbound `Cookie` header; server-side storage is the
bcrypt hash; constant-time comparison via `bcrypt.checkpw` on the
candidate walk. The raw token is never logged.
### Added
- **`device_trust` table** (`backend/migrations/017_device_trust.sql`).
Per-row id, `user_id` (FK with cascade), `device_token_hash`
(bcrypt at rest, unique index documents the no-collision
invariant), `created_at`, `expires_at` (`created_at + 30 days`),
`user_agent` (verbatim, app-layer-truncated to 1024 chars),
`last_seen_at` (refreshed on every successful lookup), `revoked_at`
(NULL means active). Secondary index on `(user_id, revoked_at)` so
the /settings list query is a covering walk.
- **`backend/app/device_trust.py`** — sibling of `otc.py` and
`passcode.py`. Carries `issue(user_id, user_agent)`,
`lookup(raw_token)`, `list_for_user(user_id)`, `revoke(user_id,
row_id)`, and `revoke_all(user_id)`. The 30-day window and the
cookie name (`rfc_device_trust`) live as module-level constants;
env-ifying them is a §19.2 candidate.
- **`§17` endpoints**
- `POST /auth/device-trust/start` — anonymous-reachable. Reads the
`rfc_device_trust` cookie; on a hit, signs the user in. On a
miss (expired, revoked, or unknown), clears the stale cookie and
returns 401.
- `GET /api/auth/me/devices` — list active trusted devices for
the signed-in user.
- `DELETE /api/auth/me/devices/{id}` — revoke a single row.
User-id scope enforced in SQL so a hostile client cannot
revoke another user's row by guessing ids.
- `DELETE /api/auth/me/devices` — revoke every active row.
- **OTC and passcode verify bodies** gain an optional
`trust_device: bool` field (default false). When true and verify
succeeds, the endpoint mints a fresh device-trust row and sets
the cookie on the response. Pre-v0.11.0 clients that omit the
field continue to behave as before.
- **Login.jsx** gains a "Trust this device for 30 days" checkbox
on both the OTC and passcode verify steps, plus a silent on-mount
call to `POST /auth/device-trust/start` so a returning user with
a valid cookie skips the email step entirely. A failure is
intentionally invisible — the user proceeds to the normal email
step.
- **`/settings/notifications` "Trusted devices" section** — lists
active rows with per-row "Revoke" + a "Revoke all devices" button
(with a `confirm()` prompt because the gesture is broad). The
surface intentionally does not single out the row whose cookie
the current request carries so a user can revoke "this device"
alongside any other from one place.
### Changed
- **`backend/app/main.py`** — the OTC and passcode verify endpoints
now also accept the `trust_device` flag and accept an injected
`Response` so they can attach the cookie. Two helpers
(`_set_device_trust_cookie`, `_clear_device_trust_cookie`) carry
the cookie attribute set in one place so the contract is
consistent across endpoints. The `Response` import is added
alongside the existing FastAPI re-exports.
- **`backend/app/api.py`** — imports `device_trust as device_trust_mod`
alongside `auth`/`db`; mounts the three `/api/auth/me/devices*`
endpoints immediately after `/api/auth/me/beta-request` so the
auth-shaped neighborhood stays clustered.
- **`frontend/src/api.js`** — exports `startDeviceTrust()`,
`listMyDevices()`, `revokeMyDevice(id)`, `revokeAllMyDevices()`.
`verifyOtc` and `verifyPasscode` accept an optional
`{ trustDevice }` argument that rides on the POST body.
### Upgrade steps (from 0.10.0)
- You **MUST** apply schema migration `017_device_trust.sql`. The
migration creates a single new table with one secondary index;
the framework runs migrations automatically at process start, so
no manual step is required beyond restarting the backend so the
migration runner picks the file up.
- You **MUST** rebuild the frontend and restart the backend after
upgrading. `frontend/package.json#version` and `VERSION` both
move to `0.11.0` and the new `Set-Cookie` shape requires the
backend to be on the matching version.
- You **MUST** serve the deployment over HTTPS. The
`rfc_device_trust` cookie is set with `Secure=True` — a
cleartext deployment will never receive the cookie back from
the browser, so the trust gesture will appear to silently fail.
Production OHM deployments already serve over HTTPS; local
development against `http://localhost` is unaffected (no cookie
is set, the OTC/passcode paths continue to work).
- You **MAY** announce the new feature to your users. Existing
signed-in sessions are unaffected — the device-trust cookie is
opt-in on the next sign-in, and a user who never checks the box
keeps the v0.10.0 behavior verbatim.
## 0.10.0 — 2026-05-28
**Minor — schema migration required; new auth path is additive.**
@@ -163,106 +595,393 @@ v0.7.0. The lockout shape (5 attempts, 15 minutes) and the length
range (420) are hard-coded in `backend/app/passcode.py`. See
§19.2 for the env-tunable candidate.
## 0.13.0 — 2026-05-28
## 0.9.0 — 2026-05-28
**Minor — schema migration required; new optional env vars.** This
release ships the cookie / privacy consent surface (roadmap item #11,
SPEC §14.5 / §14.6). Every viewer — authenticated and anonymous alike —
now sees a non-modal bottom-of-page banner on first visit asking which
categories of cookies they allow (essential / essential + analytics /
essential + analytics + other). The choice persists in `localStorage`
for anonymous viewers and in a new `cookie_consent` table for
authenticated viewers, with server-side overriding local on sign-in.
The framework also ships default `/privacy` and `/cookies` policy pages
that deployments can layer their own policy URL on top of via two new
optional env vars. No analytics SDK ships in this release — the
consent infrastructure is wired so item #13 (v0.15.0) can read from
`frontend/src/lib/consent.js` when the SDK lands.
**Minor — no schema migration; reuses v0.8.0 columns and existing
SMTP.** This release ships the admin user-management surface at
`/admin/users` and the new-beta-request notifications that feed
it (roadmap item #7, SPEC §6.1 / §15 / §17). The two halves
compose: admins receive an inbox + email signal the moment a
pending user submits the v0.8.0 capture form; clicking through
lands on the page where they Grant or Revoke access.
The page consumes the v0.8.0 schema columns
(`permission_state`, `first_name`, `last_name`,
`beta_request_reason`, `permission_decided_by`,
`permission_decided_at`) without adding new ones — migration
slot 016 stays reserved for a future release. The single new
write endpoint, `POST /api/admin/users/<id>/permission`,
replaces v0.8.0's documented manual `UPDATE users SET
permission_state='granted'` gesture with an audited UI flip.
Admin notifications ride the existing §15 chokepoint — the
`new_beta_request` event_kind is added to the enum with
category `admin-actionable`, fan-out is to every owner / admin
minus the requester themselves, and the §15.4 email dispatch
only reaches recipients whose `email_admin_actionable` toggle
is on (the default for owners + admins). The event is the
framework's first non-RFC-scoped notification — `rfc_slug` is
NULL and the email deep-link points `/admin/users` instead of
`/rfc/<slug>`.
Decision on `/admin/allowlist`: the surface stays as a
sibling sub-tab, not folded into `/admin/users`. The two have
different keys (allowlist by email pre-sign-up, user list by
user_id post-sign-up) and a union row would be confusing
rather than clarifying. The allowlist's fast-path-bypass role
from v0.8.0 is unchanged; retiring the table outright is
deferred to a later session (see §19.2).
### Upgrade steps (from 0.10.0)
1. **MAY** rebuild the frontend. The build is the same shape
as v0.10.0; the lockfile pins to `0.9.0` so `npm install`
in `frontend/` updates it cleanly. No new env vars on the
frontend; the existing `VITE_APP_NAME` requirement carries
over.
2. **MAY** restart the backend. No schema migration runs in
this release — every v0.9.0 column is from
`014_beta_access.sql` (v0.8.0). Restart only if you want
the new endpoints registered in this version's binary.
3. **SHOULD** verify SMTP can reach the deployment's admin
inbox before the first pending user submits the capture
form. v0.9.0 reuses the v0.7.0 SMTP configuration
(`SMTP_HOST`, `SMTP_PORT`, `SMTP_USER`, `SMTP_PASSWORD`,
`SMTP_STARTTLS`, `EMAIL_FROM`, `EMAIL_FROM_NAME`); a
misconfigured deployment will still write the inbox row,
but the admin won't hear about it through email. No new
env var is required — the framework reuses the existing
admin-user list (role IN ('owner', 'admin')) as the
notification recipients.
4. **SHOULD** announce the surface to existing admins. Wording
suggestion: "There's now a Users tab in /admin where you can
Grant or Revoke beta access — and you'll get an email +
inbox row when a fresh request lands. The manual SQL gesture
from v0.8.0 still works but is no longer the documented
path."
5. **MAY** drain the existing pending queue through the new UI.
If your deployment carried pending users through the v0.8.0
manual-UPDATE window, the Pending bucket on `/admin/users`
surfaces all of them with their captured profile. Granting
from the UI stamps `permission_decided_by` /
`permission_decided_at`, which any prior manual UPDATE
gestures may have left NULL (no harm done — the v0.8.0
contract didn't require those stamps).
### Added
- **Cookie consent banner** (`frontend/src/components/CookieConsentBanner.jsx`).
Non-modal, bottom of viewport. Three single-select choices with
inline descriptions. Visible until the user makes a choice; hides
thereafter. Reachable for revision via the settings surface.
- **Consent helper** (`frontend/src/lib/consent.js`). Exports
`getConsent()`, `hasChosen()`, `onConsentChange(cb)`, `setConsent()`,
`hydrateFromServer()`, `clearLocal()`. Cross-tab sync via the
`storage` event. Item #13's analytics SDK reads consent here before
importing.
- **Privacy and cookies policy pages**
(`frontend/src/pages/Privacy.jsx`, `frontend/src/pages/Cookies.jsx`).
Default minimal policies that describe the framework's stance and
list the cookies the framework sets. Deployments override via the
two new env vars below; the framework's stub always renders above
the link so the framework-level contract stays visible.
- **"Privacy & cookies" tab** in `/settings/notifications` showing
the current consent choice, the recorded-at stamp, and a "Change"
button that re-opens the banner via a custom DOM event.
- **`§17` endpoints** —
- `GET /api/users/me/cookie-consent` — read the current consent
record.
- `PUT /api/users/me/cookie-consent` — write a new consent record.
Upserts a single row per user, stamps `recorded_at` to now,
accepts `essential` for symmetry but always persists it as true.
- **Schema migration** `013_cookie_consent.sql` — new
`cookie_consent` table keyed by `user_id`, three flags
(`essential`, `analytics`, `other_cookies`), and `recorded_at`.
(Renumbered from `012_*` during driver integration because v0.7.0
also added a `012_otc.sql` migration that landed in the integration
order before this one.)
- **SPEC `§14.5` Cookie / privacy consent** — settles the banner
shape, the three-category single-select, the storage shape (local
for anon, server row for authenticated), the precedence rule on
sign-in, and the `consent.js` helper surface for downstream
callers including item #13.
- **SPEC `§14.6` Privacy and cookies policy pages** — settles the
`/privacy` and `/cookies` routes, the framework's stub content, and
the `VITE_PRIVACY_POLICY_URL` / `VITE_COOKIES_POLICY_URL` override
shape.
- **SPEC `§5`** — names the `cookie_consent` table in the canonical
app-tables list.
- **SPEC `§17`** — lists the two new cookie-consent endpoints.
- **SPEC `§19.2`** — surfaces four candidates: policy content via
content-repo file vs env var, GPC / DNT headers, multi-language
consent text, and the item #13 analytics-SDK gating dependency.
- **`POST /api/admin/users/<id>/permission`** — body
`{state: 'pending'|'granted'|'revoked'}`. Flips the column,
stamps `permission_decided_by` + `permission_decided_at`,
and writes a `permission_events` row with event_kind in
`{permission_granted, permission_revoked,
permission_repended}`. Refuses 422 on self-flip (symmetric
to `set_mute` / `set_role` self-action refusals) and 422
on invalid state. Returns `{ok, permission_state, changed}`
where `changed=false` indicates a no-op (the requested
state already matched).
- **Widened `GET /api/admin/users` response** carrying
`permission_state`, `first_name`, `last_name`,
`beta_request_reason`, `created_at`,
`permission_decided_at`, plus joined
`permission_decided_by_login` /
`permission_decided_by_display`. Sort order surfaces
`pending` rows first (the admin queue), then `granted`,
then `revoked`; within a bucket, owners precede admins
precede contributors, with recency as the tiebreaker.
- **`new_beta_request` event_kind** in the §15.1 enum. Fired
by `notify.fan_out_new_beta_request` from the first
successful `POST /api/auth/me/beta-request` (re-submits
from the same pending user don't re-fire — the row's
first-time-complete check guards against carpet-bombing).
Recipients: every owner + admin minus the requester
themselves. Category: `admin-actionable`. Deep-link:
`/admin/users`. Actor: the requester per §15.9.
- **`/admin/users` page enhancements** in `Admin.jsx`. The
Users tab gains a state filter chip row (All / Pending /
Granted / Revoked with counts), a Grant / Revoke control
column, a Permission state badge, and an expandable
"why they want access" row beneath each pending user.
Sign-up timestamp surfaces in a new column.
- **`frontend/src/api.js#setUserPermission`** — client for
the new endpoint, neighboring `setUserMute` and
`setUserRole`.
- **`/beta-pending` copy update** in `BetaPending.jsx`. The
"your request is in review" page now honestly references
the admin-email signal v0.9.0 ships and admits the
framework does not commit to an SLA — turnaround depends
on operator availability, and the deployment operator is
the right person to ask if a wait runs long. No
deployment-specific text is baked in; per-deployment copy
lives in §13 of the deployment's repo, not in the
framework.
- **`backend/tests/test_admin_users_vertical.py`** — 10 new
tests covering: a beta-request submission fans
notifications to every admin + owner (and not to the
requester or to contributors); the requester's profile
fields land in the notification payload; re-submitting
the capture form does not re-fan; the `admin-actionable`
category mapping is wired; the `/api/admin/users`
listing carries the v0.8.0 columns with `pending` rows
sorted first; the permission-flip endpoint promotes
pending → granted with the right audit shape; the
endpoint promotes granted → revoked; the endpoint
refuses self-flip with 422; the endpoint refuses
non-admin callers with 403 and anonymous callers with
401; the endpoint refuses invalid states with 422; a
state-already-matches flip returns `changed=false`
without writing an audit row.
### Changed
- **`frontend/.env.example`** — documents the two new optional env
vars `VITE_PRIVACY_POLICY_URL` and `VITE_COOKIES_POLICY_URL`. Unset
is supported; defaults render the framework's stub.
- **`backend/app/api_notifications.py`** — module docstring grew two
endpoint lines; the new endpoints sit alongside the existing
`/api/users/me/*` neighbors.
- **`frontend/src/App.jsx`** — registers `/privacy` and `/cookies`
routes (anonymous-reachable), wires `<CookieConsentBanner>` into
the global chrome, and listens for a `rfc-app:cookie-consent-reopen`
custom event to re-open the banner from the settings surface.
- **`backend/app/api.py`** — the `POST /api/auth/me/beta-request`
handler now calls `notify.fan_out_new_beta_request` after the
capture UPDATE lands, gated on the row not previously having
all three profile fields populated (so re-submits don't
re-fan).
- **`backend/app/api_admin.py`** — the `list_users` query joins
against `users d ON d.id = u.permission_decided_by` for the
deciding-admin handle. The new `set_permission` endpoint
lives alongside `set_role` / `set_mute`.
- **`backend/app/notify.py`** — adds
`CATEGORY_ADMIN_ACTIONABLE`, the `fan_out_new_beta_request`
helper, and the `new_beta_request` arm in `render_summary`.
- **`backend/app/email.py`** — the `_EVENT_TO_CATEGORY` map
carries `new_beta_request → admin-actionable`, and
`_deep_link` routes framework-scoped admin signals to
`/admin/users` instead of `/rfc/<slug>`.
- **`frontend/src/components/Admin.jsx`** — the `UsersTab`
component is rewritten with state filter chips, a per-row
`UserRow` + `PermissionCell` decomposition, and a
`useMemo`-cached counts table. The pending-row reason
blockquote renders as a secondary `<tr>` beneath the user
row when present.
- **`frontend/src/components/BetaPending.jsx`** — pending-state
copy revised to reference the admin email signal honestly.
- **`frontend/src/App.css`** — new admin-chip / permission-badge
/ user-row-reason rules.
- **`SPEC.md`** §6 opening, §15.1 event-kinds enum, §17 admin
endpoints, §19.2 candidates list — per §19.3 rule-2.
- **`VERSION`** → `0.9.0`. `frontend/package.json#version` and
the lockfile mirror.
### Upgrade steps (from 0.7.0)
### Environment variables
- You **MUST** rebuild the frontend and restart the backend after
upgrading. `frontend/package.json#version` and `VERSION` both move
to `0.13.0` and the build embeds the new env-var contract.
- You **MUST** apply schema migration `013_cookie_consent.sql`. The
migration creates a single new table keyed by `user_id` with three
flag columns and a `recorded_at` stamp. The framework runs
migrations automatically at process start; no manual step is
required beyond restarting the backend so the migration runner
picks the file up.
- You **MAY** set `VITE_PRIVACY_POLICY_URL` to an http(s) URL that
points at your deployment's full privacy policy. The framework's
`/privacy` page renders its built-in stub above a link to the
configured URL. Unset is supported — the stub is sufficient for a
default-config deployment.
- You **MAY** set `VITE_COOKIES_POLICY_URL` to an http(s) URL that
points at your deployment's full cookies policy. Same shape as the
privacy URL.
- You **MAY** announce the new consent banner to your users. Existing
authenticated users will see the banner on their next visit
(because their `cookie_consent` row does not yet exist); their
current sessions remain valid.
None new. The release reuses the v0.7.0 SMTP configuration
(`SMTP_HOST`, `SMTP_PORT`, `SMTP_USER`, `SMTP_PASSWORD`,
`SMTP_STARTTLS`, `EMAIL_FROM`, `EMAIL_FROM_NAME`,
`EMAIL_ENABLED`, `EMAIL_BUNDLE_THRESHOLD`) and the existing
admin-user list (role IN ('owner', 'admin')) as the
notification recipients. A deployment whose SMTP is misconfigured
will still see the inbox rows; only the email channel is muted.
### Deferred to later releases
- **Grant / revoke notification to the user** — symmetric
signal: when an admin grants or revokes access, fire a
`personal-direct` notification (event_kind
`permission_change_affecting_me`, already in the §15.1
enum) so the affected user sees the state change in
their inbox and email. v0.9.0 audits the gesture in
`permission_events` but does not yet escape the
app-internal log to the user. See §19.2.
- **Decline-with-reason on revoke** — the current Revoke
gesture takes only a confirmation; a follow-up release
may capture a free-text reason in
`permission_events.details`. See §19.2.
- **Allowlist deprecation** — the `/admin/allowlist` sub-tab
stays in place in v0.9.0 (the two surfaces have different
keys and a union row would be confusing). Retiring the
table outright is deferred to a session that can re-read
the post-v0.9.0 operator experience and decide whether
the fast-path-bypass role is still pulling weight. See
§19.2.
## 0.8.0 — 2026-05-28
**Minor — schema migration required; admission semantics shift.**
This release replaces the v0.3.0 / v0.7.0 `allowed_emails` admission
gate with an admin-grant flow (roadmap item #6, SPEC §6.1 / §6.2 /
§14.1 / §17). Anyone with a valid email can sign in via the v0.7.0
OTC flow; the OTC request endpoint no longer consults the
allowlist. A fresh user lands in `permission_state='pending'` until
an admin grants access. The first-OTC sign-in captures first name,
last name, and a free-text "why I should be included in the beta"
via a new `POST /api/auth/me/beta-request` endpoint; the captured
fields populate the same `users` row alongside the OAuth-era
columns.
A pending user has the same read access an anonymous viewer has —
the catalog, RFC bodies, the philosophy page, and every public
conversation are reachable. Every write-shaped endpoint
(`auth.require_contributor` floor) refuses pending users with 403.
The frontend renders a thin "Your beta access request is in
review" banner on every page and re-purposes the v0.3.0
`/beta-pending` page as the post-capture landing surface.
Grandfathered behavior: every `users` row at migration time
carries `permission_state='granted'` via the column default, so
existing contributors are unaffected. The OAuth fallback at
`/auth/callback` still consults the v0.3.0 allowlist (legacy
path); the OTC flow does not.
The `allowed_emails` table stays in the schema as a fast-path
bypass — the v0.3.0 admin UI continues to manage it, but the OTC
request handler no longer reads it. v0.9.0 (roadmap item #7) is
expected to ship the admin user-management page that replaces the
allowlist surface entirely; until then, admin grants are done by
direct DB `UPDATE`.
### Upgrade steps (from 0.13.0)
1. **MUST** restart the backend so migration `014_beta_access.sql`
runs. The migration adds `permission_state` (default `'granted'`,
so existing rows pass through unaffected), `first_name`,
`last_name`, `beta_request_reason`, `permission_decided_by`,
and `permission_decided_at` to the `users` table, plus an index
on `permission_state` for the pending queue. The migration is
ALTER-TABLE-based (no table rebuild) — every foreign key and
existing row passes through untouched.
2. **MUST** rebuild the frontend. The `Login.jsx` surface now
runs a conditional third step (the capture form) on fresh OTC
sign-ins; `BetaPending.jsx` carries the new "your request is
in review" copy; `App.jsx` renders a thin pending-access
banner. The build embeds the new `/api/auth/me/beta-request`
client call.
3. **SHOULD** announce the new admission flow to existing users.
Wording suggestion: "We've replaced our email-allowlist gate
with an admin-review flow. Existing users are unaffected;
new visitors sign in with their email, tell us a bit about
themselves, and an admin reviews their request before
discussion and contribution unlock." Existing sessions
remain valid.
4. **SHOULD** plan the admin grant mechanism. v0.8.0 does not
ship a UI for the grant — v0.9.0 (roadmap item #7) will. For
the v0.8.0 window, an admin grants access via direct DB
gesture:
```sql
UPDATE users
SET permission_state = 'granted',
permission_decided_by = <admin_user_id>,
permission_decided_at = datetime('now')
WHERE email = '<approved>';
```
The pending queue lives in `SELECT * FROM users WHERE
permission_state = 'pending' ORDER BY created_at`.
5. **MUST** decide whether to drain the `allowed_emails` table.
The OTC request handler no longer consults it; populated
rows are inert at the request surface. Three operator
choices, all valid:
* **Leave as-is** (the framework's default behavior — the
v0.3.0 admin UI continues to work, the rows stay as a
fast-path bypass record). Recommended if you anticipate
v0.9.0's user-management page folding the allowlist UI
into its surface.
* **Drain via the existing admin UI** (`/admin/allowlist`)
— one row at a time, no data loss elsewhere.
* **Bulk-drain via DB** — `DELETE FROM allowed_emails;`
drops every row; the table stays.
6. **MAY** announce write access individually to grandfathered
users you want to keep at `'granted'`. The default-`'granted'`
migration means no action is required for them; this step
exists only if you want to send a "you're still in" message.
### Added
- **`backend/migrations/014_beta_access.sql`** — adds
`permission_state` (CHECK in `('pending', 'granted', 'revoked')`,
default `'granted'`), `first_name`, `last_name`,
`beta_request_reason`, `permission_decided_by` (FK to users,
ON DELETE SET NULL), `permission_decided_at` to the `users`
table. Plus `idx_users_permission_state` for the pending
queue.
- **`POST /api/auth/me/beta-request`** — body
`{first_name, last_name, beta_request_reason}` (all required;
bounds 120 / 120 / 4000). Writes the fields to the signed-in
user's row and leaves `permission_state='pending'`. Refuses
HTTP 409 for already-granted / revoked users; refuses HTTP
401 for anonymous callers.
- **`needs_profile` flag** on the `/auth/otc/verify` response.
`true` iff the user is `permission_state='pending'` AND
carries no profile fields yet (a fresh OTC sign-in). The
Login.jsx surface uses the flag to gate the capture step.
- **`permission_state` field** on the `/api/auth/me` response,
plus `first_name`, `last_name`, `beta_request_reason`, and
the same `needs_profile` flag.
- **First-OTC profile capture step** in `Login.jsx`. Third
step in the sign-in surface, gated by the verify response's
`needs_profile` flag.
- **`/beta-pending` repurpose** in `BetaPending.jsx`. The
page now reads as "your request is in review" when the
viewer is pending; the v0.3.0 "private beta" framing
remains as the anonymous-viewer fallback.
- **Thin pending-access banner** at the top of every page
for `permission_state='pending'` viewers (other than
`/beta-pending` itself).
- **SPEC `§6` opening / `§6.1` / `§6.2` / `§14.1` / `§17` /
`§19.2`** corrections per §19.3 rule-2 — the admission
shift, the orthogonality of permission_state vs role / muted
/ notification-mutes, the new endpoints, and the
newly-surfaced §19.2 candidates (admin user-management page,
allowlist deprecation, admin notification on new request).
- **`backend/tests/test_beta_access_vertical.py`** — 9 new
tests covering: a fresh OTC user lands pending with empty
profile; the capture endpoint populates the fields and
keeps state pending; the capture endpoint refuses
anonymous / granted / revoked callers; a pending user is
refused write endpoints; an admin grant promotes pending →
granted; a grandfathered user is unaffected by the
migration; the OTC request endpoint accepts any email
regardless of allowlist state; the `allowed_emails` table
is still present in the schema.
### Changed
- **`backend/app/auth.py#require_contributor`** widens its
gate to refuse `permission_state != 'granted'` with HTTP
403. The §6.1 contributor capabilities (propose, branch,
PR, chat, claim) all funnel through this dependency, so
the widening covers them transitively. `SessionUser` now
carries `permission_state` (default `'granted'` for the
dataclass-default fallback path).
- **`backend/app/otc.py#request_code`** drops the allowlist
check from the OTC request flow. The `RequestOutcome`
shape loses the `'allowlist'` reason (replaced by
`'sent'` / `'cooldown'` / `'invalid'`).
- **`backend/app/otc.py#provision_or_link_user`** sets
`permission_state='pending'` explicitly on a fresh row.
Grandfathered (link-by-email) users pass through with
their existing column value.
- **`backend/app/auth.py#provision_user`** (OAuth fallback)
now sets `permission_state='granted'` explicitly on a
fresh row. The OAuth callback still consults the
`is_allowed_sign_in` allowlist check (the legacy fallback
path retains its v0.3.0 admission shape during the OAuth
migration window).
- **`backend/tests/test_otc_vertical.py`** — the
`test_otc_request_silently_drops_when_email_not_on_allowlist`
test (asserted the v0.7.0 allowlist gate) is replaced by
`test_otc_request_admits_emails_regardless_of_allowlist_population`
which asserts the v0.8.0 open-request contract. The
on-list test stays as a regression net for the
rate-limit / outbound-buffer plumbing.
- **`SPEC.md`** §6 opening, §6.1, §6.2, §14.1, §17, §19.2
per §19.3 rule-2.
- **`VERSION`** → `0.8.0`. `frontend/package.json#version` and
the lockfile mirror.
### Deferred to later releases
- **Admin user-management page** at `/admin/users` (item #7,
v0.9.0) — replaces the manual DB `UPDATE` gesture.
- **Allowlist UI deprecation** (also v0.9.0) — once the
admin user-management page lands, the `/admin/allowlist`
surface and the `allowed_emails` table both retire.
- **Admin email notification on new beta request** (item #7
again, v0.9.0).
- **Revoke gesture in the UI** — the `permission_state='revoked'`
state is wired in the schema and the auth gate; v0.9.0 ships
the admin UI that flips the column.
## 0.7.0 — 2026-05-28
@@ -851,7 +1570,3 @@ names itself.
- `CLAUDE.md` at the repo root capturing the separation-of-concerns
rule for working sessions.
## 0.1.0 — v1 build
Initial release. See `docs/DEV.md` for the slicing plan and build
history.
+605
View File
@@ -0,0 +1,605 @@
# Using the RFC app
This is the user-facing guide to the Wiggleverse RFC framework — how to
read what's here, propose a new RFC, contribute to one that already
exists, and understand who is allowed to do what.
This guide describes the framework. Individual deployments brand and
configure themselves independently — the name in the header and the
corpus the RFCs are about belong to the deployment, not to this
document.
For the *why* of the framework, read the [philosophy](/philosophy).
For the binding technical contract, see `SPEC.md` in the repository.
---
## Reading without signing in
You can read the catalog and every public RFC without an account.
Anonymous visitors can:
- Browse the catalog of super-drafts and active RFCs.
- Open any RFC and read its canonical body.
- Read any public branch — its diff and its chat thread.
- Read any pull request — its diff, its conversation, its review
comments.
- Read the discussion that has accumulated on an RFC's main view.
Reading is open by design. The framework's claim is that the *argument
behind a definition* is the evidence that the definition was earned,
and an argument that disappears behind a sign-in wall stops carrying
that evidence.
What you cannot do without an account: chat, propose a new RFC,
create a branch, open a PR, drop a flag, or post on a discussion
thread. Every write affordance is replaced with a sign-in prompt.
---
## Signing in
While the framework is in private beta, only invited email addresses
can complete sign-in. If your email is on the allowlist, the
"Sign in" button in the header completes the flow and lands you on
the catalog with full read and write access. If your email is not on
the allowlist, you'll be sent to a short "pending" page explaining
the gate.
Once you have an account, you're a **contributor** by default — the
role that grants every write affordance the app exposes, scoped by
the per-RFC and per-branch rules described below.
---
## Proposing a new RFC
A new RFC begins as a proposal. The "+ Propose new RFC" button at
the bottom of the catalog opens a small modal that collects four
things:
- **Title.** The word, concept, or topic this RFC would define.
- **Slug.** A kebab-cased identifier derived from the title. It is
the entry's stable handle from this moment until it graduates;
collisions with existing entries or open proposals are caught
inline.
- **Pitch.** One or two paragraphs answering *why this RFC is
needed*. This becomes the body of the entry.
- **Tags.** Optional. The AI suggests tags from the pitch; you can
accept, dismiss, or type your own.
Submitting the modal does one concrete thing: it opens a pull
request against the framework's meta repository, adding one new
file under `rfcs/`. There is no other Git artifact and no other
side-effect. You are returned to the **pending-idea view** for the
new proposal.
A pending idea is publicly readable but not yet a super-draft. The
catalog surfaces it in a "Pending ideas" disclosure at the bottom
of the list. A conversation can accumulate on the pending-idea view
before it is admitted — contributors can argue, in public, about
whether the entry belongs in the catalog at all.
Three outcomes are possible:
- **Merge.** An admin or owner merges the proposal PR. The entry
becomes a super-draft and graduates from the "Pending ideas"
section into the main catalog. Any conversation that accumulated
on the pending-idea view migrates with it.
- **Decline.** An admin or owner declines, attaching a written
comment. You see the comment on your next visit, along with a
one-click affordance to revise and re-propose.
- **Withdraw.** You can withdraw your own proposal at any time. The
entry will not appear in any default view; the conversation that
accumulated stays attached to the closed PR as historical record.
You are automatically the first owner of any RFC you propose. The
claim flow described under [Roles & permissions](#roles--permissions)
is for *other* contributors to add themselves as owners later, not
for the proposer.
---
## What a super-draft is
A super-draft is an entry that has been admitted to the catalog but
does not yet have its own dedicated repository. Most of the
argument that shapes a definition happens here. The framework
assumes — and the philosophy explicitly invites — that many
super-drafts will not survive the argument, and that is fine. The
entries that do survive earn their place in the catalog by being
defensible in public.
Opening a super-draft from the catalog gives you the same surface
an active RFC uses:
- The canonical body in the centre, read-only by default.
- A chat thread on the right where the public conversation lives.
- A breadcrumb dropdown listing any in-flight edit branches and
any open body-edit PRs against this entry.
- A "Start Contributing" affordance that cuts a fresh edit branch
and lands you in contribute mode.
Edits to a super-draft body propagate through pull requests against
the meta repository — there is no dedicated RFC repository yet.
---
## What an active RFC is
An active RFC is an entry that has been **graduated**. It has its
own dedicated repository, an integer `RFC-NNNN` identifier, and a
canonical body file (`RFC.md`) inside that repository. The catalog
distinguishes super-drafts and active RFCs at a glance.
Opening an active RFC gives you:
- `main` — the canonical body, always read-only. Changes to `main`
arrive exclusively through pull requests.
- A breadcrumb listing every open branch and pull request on this
RFC.
- A per-branch chat thread on the right. Each branch has its own
conversation, including `main` itself.
- A "Start Contributing" affordance: on `main` it cuts a new branch
and lands you on it in contribute mode; on any other branch you
already have push access to, it flips that branch into
contribute mode.
---
## Discussion vs contribution
The framework draws an explicit distinction between two surfaces
that other tools tend to conflate:
- **Discussion** is what the RFC is *for*. The chat thread on an
RFC's main view is the place for "what about this part?" or
"have we considered…?" questions that don't yet warrant proposing
a specific edit. Posting on a discussion thread does not create
any Git artifact; the conversation lives in the app database.
- **Contribution** is how an RFC *changes*. Editing the canonical
body requires opening a branch and, eventually, a pull request.
The pull request is the place a specific proposed change is
reviewed and merged.
Reading both surfaces is open to anonymous visitors. Posting on
either requires a contributor account.
---
## Working on a branch
Contribute mode flips one branch into edit-enabled. The centre
column splits: a markdown source pane on the left, a live-rendered
preview on the right. Fenced `mermaid` blocks render as diagrams in
the preview.
Two kinds of edits accumulate on a branch:
- **AI-proposed changes.** You ask the AI a question or request a
revision in the branch's chat. When the AI proposes a concrete
edit, that edit appears as a *change card* in a panel below the
chat — not yet applied to the document. You can **accept**,
**decline**, or **edit before accepting**. Accepting produces
one commit on the branch with the original text, the proposed
text, and the AI's reason recorded in the commit body.
- **Manual edits.** Typing directly into the source pane buffers
locally and flushes as a single commit on an idle window, a
branch switch, or an explicit "Save now" button. Manual edits
also appear as change cards in the same panel — same evidence
shape, different author.
Every accepted change is one commit. The framework does not
support squash-merges or fixup-style cleanups: the per-change
commit granularity is the framework's evidence unit, and
collapsing it would erase what was earned.
### Discuss mode vs contribute mode
A branch defaults to discuss mode — read-only, with chat enabled.
AI proposals still appear in chat, but they are *buffered* rather
than applied; a single CTA invites you to flip the branch into
contribute mode if you want to act on them. The toggle is an
*intent* affordance, not a permission one. If you don't have push
access to the branch, the toggle is disabled with a sign-in or
request-access path.
`main` is special: contribute mode is never available there. The
"Start Contributing" button on `main` always cuts a new branch.
### Flags
Anywhere you can read, you can drop a flag. A flag is the
lightweight "I'm pointing at this, it's a problem" gesture — a
single short declarative statement anchored to a passage. Creating
a flag requires a contributor account but does not require push
access to the branch: any signed-in contributor who can read a
passage can point at it and say it's wrong.
Flags don't block PR merges by design — making them a merge gate
would re-create the failure mode where contributors hastily "resolve"
threads to unblock a button. Flags are prominent on PR headers but
non-blocking.
### Branch visibility
A new branch is publicly readable by default. The branch creator
can flip a branch to private, in which case only the creator, any
explicit grantees, and the RFC's per-RFC owners and arbiters can
read it. Owners and arbiters can flip it back.
**Opening a PR makes the branch fully public.** If your branch is
currently private, the "Open PR" affordance asks you to confirm
this before submitting. There is no concept of a private PR — the
framework's evidence claim depends on the argument being readable.
### Who can push to a branch
Every branch has one of three contribute modes:
- **`just-me`** (default) — only the branch creator can push.
- **`specific`** — only the branch creator and explicitly granted
contributors can push.
- **`any-contributor`** — any signed-in contributor can push.
The branch creator and the RFC's per-RFC owners and arbiters can
change this setting at any time.
### Branch hygiene
A branch with no associated PR auto-closes after 30 days of
inactivity. A closed branch is deleted from the Git host 60 days
later. Closed branches remain in the catalog under a "show closed"
filter — closing is a state, not a censorship event. The chat
attached to a closed or deleted branch is preserved as historical
record.
Owners and arbiters can *pin* a branch to disable the auto-close
timer if the work is paused but legitimately ongoing.
---
## Opening and reviewing a pull request
A pull request is the deliberate "ready for review" gesture for
work that has accumulated on a branch. The "Open PR" affordance is
available on any branch with at least one commit ahead of `main`.
The PR creation modal collects two AI-drafted fields, both editable
before submit:
- **Title.** A one-line description of the change, in spec voice.
- **Description.** Two to four sentences pulling from the branch
chat, written for an arbiter.
There is no reviewer picker. The RFC's arbiters are the implicit
reviewer set.
### The PR review page
The review page shows the diff, the branch's compressed chat
(messages that produced accepted changes are expanded, the rest is
behind a "Show full conversation" toggle), and the review-comment
surface inline below the chat.
Review comments are not a separate concept from chat — they live in
the same thread, anchored to a range in the diff. The framework's
claim is that the disagreement an arbiter raises about a proposed
change is the same *kind* of thing as the disagreement that
produced the proposed change in the first place, and the two should
share a surface.
Each PR records a per-user seen-cursor. New diff hunks and new
conversation messages since your last visit render with a subtle
accent. The cursor advances on view; you do not have to mark
anything as read.
### Merging a PR
Per-RFC owners and arbiters can merge; app-wide admins and owners
also retain this capability. The merge produces a no-fast-forward
commit on `main`, preserving every per-acceptance commit as an
individually reachable node in `main`'s history.
Merge is hard-blocked **only** by Git-level conflicts with `main`.
Open review threads, pending change-cards, unresolved chat threads,
and open flags do not block merge by design.
### Conflicts with main
A conflict surfaces on the PR page as a read-only banner. A "Start
resolution branch" affordance cuts a fresh branch off `main`'s
current tip, replays the work into it (asking the AI to resolve
unambiguous conflicts, surfacing the rest for you), and opens a new
PR. The original PR auto-closes when the resolution PR merges.
Fixup commits on the existing branch are not supported. Per-change
commit granularity is the framework's evidence unit; admitting
"fix merge conflict with main" commits would dilute it.
---
## Graduation: super-draft → active RFC
Graduation is the moment a super-draft becomes a canonical entry
in the catalog. It is initiated by an app-wide admin, an app-wide
owner, or one of the RFC's per-RFC owners or arbiters from the
super-draft's page.
Two preconditions block the action:
- **The super-draft must have at least one owner.** The proposer
is automatically the first owner; if they have stepped away, any
contributor can use the "Claim ownership" affordance to add
themselves.
- **No open body-edit PRs against the super-draft's entry.** An
open body-edit PR would attempt to re-introduce a body to a
frontmatter-only entry after graduation runs. Merge or withdraw
them first.
When the dialog confirms, the framework runs a transactional
sequence: create a fresh Git repository for the RFC, seed it with
the super-draft's body as `RFC.md`, update the meta-repo entry to
`state: active` with the integer ID and the new repository's URL,
auto-merge that update. If any step fails partway, the sequence
rolls back — the half-created repository is deleted and the
unmerged update is abandoned. The dialog shows each step in flight
and tells you exactly what happened.
The chat thread on the super-draft moves to the new repository's
`main` chat at graduation. Edit-branch chats from the super-draft
phase stay attached to their original branches on the meta repo
and surface from the new RFC view under a "Pre-graduation history"
section.
Graduation is not reversible. The path forward from an active RFC
is withdrawal, not back to super-draft.
---
## Withdrawing and reopening
An active RFC or a super-draft can be withdrawn by the proposer
(for a super-draft they proposed) or by an admin or owner. A
withdrawn entry stays in the catalog as a historical record but is
hidden from default views. The entry is filterable back in.
An admin or owner can reopen a withdrawn entry back into the
super-draft state. The history is preserved across the transition.
---
## AI in the chat
The chat on every RFC, super-draft, branch, and PR has an AI
participant by default. The framework treats the AI as one voice
among many in a public argument — not an oracle, and not a
co-author whose name lands on commits.
You invoke the AI by writing into the chat composer and submitting.
Each message can pick a model from the picker (the option list is
configurable per RFC). The AI responds in the chat; when its
response includes a concrete change to the document, that change
appears as a card you can accept, decline, or edit.
When you accept an AI's proposed change, the commit's
`On-behalf-of:` trailer names *you*, not the AI. The AI's authorship
survives only as evidence — the original proposal in the commit body
and the message that produced it in the chat record. The framework
is explicit about this: AI participation produces evidence; it does
not produce authorship.
Two configuration knobs scope AI participation per RFC:
- **Which models are available.** The meta-repo entry's frontmatter
carries an optional `models:` list. Absent means the RFC inherits
whatever models the deployment is provisioned to run. An empty
list (`models: []`) opts the RFC out of AI entirely — every AI
surface is absent rather than disabled-but-present.
- **Whose credentials pay.** By default the deployment operator's
API credentials cover AI calls on every RFC. A `funder:`
frontmatter field can name a single contributor whose registered
credentials pay for AI calls on this RFC instead. The named
contributor must explicitly consent from their settings page;
either side can revoke at any time.
Per-RFC AI configuration is edited through the meta-repo PR flow
that governs the rest of the entry's frontmatter — by the RFC's
per-RFC owners and arbiters, or by app-wide admins or owners.
---
## Notifications
The framework's public-async work model produces signals that
shouldn't all reach you the same way. Five surfaces compose:
- **In-app inbox.** The durable triage surface. One mental space
across every RFC you have any relationship to, with per-RFC and
per-category filters. Reachable from the inbox icon in the
header.
- **Badges.** Ambient pull-ins. A single integer beside the inbox
icon (count of unread notifications). A small binary dot on
individual catalog rows for watched RFCs with unseen activity.
No per-row counts and no per-section counts.
- **Toasts.** Transient mid-session signals. Used only for your own
actions completing, and for events arriving on the view you're
currently looking at.
- **Email.** The single channel that escapes the app. Opt-in per
category, conservative defaults. One-click unsubscribe per
category.
- **Digest.** Aggregation for activity on watched RFCs you haven't
triaged through any other channel.
### Watch states
Every RFC has one of three implicit relationship states for you:
- **Watching.** You receive structural signals for the RFC.
- **Following.** You receive only churn-grade signals (new
commits, new chat messages on threads you didn't participate
in). This is a lighter relationship than watching.
- **Muted.** You receive no signals for the RFC. The mute is
per-RFC and self-imposed; it does not affect what others see
or what reaches you on *other* RFCs.
Watch states transition automatically based on your participation,
with explicit overrides available from each RFC's header and from
the notification settings page.
### Email categories
Four categories with distinct defaults:
- **Personal-direct events** — default on. Signals where you are
the named subject. The contract is that when your name is on the
action, the framework reaches out of band.
- **Watched-RFC structural events** — default off. PR opened on a
watched RFC, PR merged, graduation, withdrawal. Inbox and badges
carry these by default; the email toggle is opt-in.
- **Watched-RFC churn** — permanently off, by design. Per-commit
and per-message email is intentionally not offered. The digest
aggregates this activity weekly.
- **Admin-actionable events** — default on for admins and owners,
unused for contributors.
### Quiet hours
You can set a daily window during which email notifications are
held. Messages held during the window are released at window end —
bundled into a single "Activity while you were away" email if a
threshold accumulated, otherwise sent individually.
---
## Roles & permissions
Authorization in this framework is owned by the app itself, not by
the Git host. The Git host sees only a single bot account — every
commit, every PR, every merge passes through it on a user's behalf
— and the *app* decides which users are authorized to ask the bot
to do which things.
### The four app-wide roles
Each role is a strict superset of the one below it.
1. **Anonymous.** Anyone who has not signed in. Can read public
RFCs, public branches, and public PRs; cannot chat, propose,
create branches, or open PRs.
2. **Contributor.** The default role for any authenticated
account. Adds everything anonymous can do, plus: propose new
RFCs, create branches on any RFC repository, open PRs from
branches they have push access to, post on chat anywhere they
can read, claim ownership of unclaimed super-drafts.
3. **Admin.** Adds the ability to act on any RFC, anywhere in the
framework. Concretely: merge any PR on any RFC, graduate any
super-draft, set branch visibility on anyone's behalf, withdraw
or reopen any entry, write-mute or restore any contributor,
grant or revoke the **admin** role.
4. **Owner.** Adds two capabilities admin does not have: grant or
revoke the **owner** role itself, and disable an account
entirely. The framework names a single "owner zero" at
bootstrap.
The practical difference between admin and owner is narrow but
load-bearing: admin is the operational tier — it does the day-to-
day moderation and stewardship work; owner is the tier that
controls the admin tier. Disabling an account and creating other
owners are owner-only because they affect the framework's chain of
authority itself.
The app refuses to let the last owner demote themselves silently —
losing the last owner would leave nobody able to grant the role
back. Role changes are recorded in an append-only `permission_events`
log; an admin's own admin/users page shows the log of who promoted,
demoted, or muted whom.
### Per-RFC delegated authority
The four roles above are framework-wide. Within an individual RFC,
the meta-repo entry's frontmatter names two additional groups:
- **`owners:`** — contributors elevated for this RFC. They can
grant push access on any branch in the RFC, merge any PR on the
RFC, change branch visibility, and withdraw the RFC.
- **`arbiters:`** — contributors with merge authority for this RFC.
Functionally similar to per-RFC owners for merge decisions; the
distinction matters in some configuration paths.
Per-RFC owners and arbiters are **not** app-wide admins. Their
elevated powers are scoped strictly to the RFC named in the
frontmatter. This is what lets the framework distribute work
without putting one person on the hook for every action.
The proposer of an RFC is automatically the first per-RFC owner.
Additional per-RFC owners are added through a "Claim ownership"
PR against the meta repository; app-wide admins or owners merge
it.
### Per-branch contribute grants
Within an RFC, the branch creator and the RFC's per-RFC owners
and arbiters can grant push access to specific contributors on a
specific branch — `specific` contribute mode, described under
"Working on a branch."
### The write-mute
An app-wide admin or owner can **mute** a contributor. A muted
account retains read access and keeps its existing branches, but
cannot create new branches, open new PRs, propose new RFCs, or
post chat. This is a moderation tool, distinct from removing the
account; restoring is the reverse gesture.
The write-mute applies only to contributors. Promoting a user to
admin or owner is the way to remove a user's write-restriction in
the structural sense; the write-mute is for *retaining* an account
while removing its ability to act.
Every mute and every restore is recorded in `permission_events`.
### Three different "mutes"
The word "mute" appears in three structurally distinct places.
They share a word and nothing else.
- **Write-mute.** Admin-imposed. Removes a contributor's ability
to post or push. Described above.
- **Per-RFC notification mute.** Self-imposed. Sets your watch
state on a specific RFC to *muted* — you stop receiving signals
for that RFC, in inbox, badges, and email. Does not affect what
others see.
- **Per-user notification mute.** Self-imposed. Suppresses
notifications produced by a specific other user, anywhere in
the framework. Notification-volume only — it does not affect
what you can read.
A write-muted contributor continues to receive notifications
normally, so they can triage what they can't act on, and so a
restore lands cleanly.
### Audit trail
Every gesture that changes app state — role changes, mutes,
graduations, withdrawals, grant changes — is recorded in
append-only logs the app maintains. Git commit history is for
code archaeology; the app's audit log is the accountability
record. An admin's page surfaces both `permission_events` (the
role/mute log) and `actions` (the state-transition log) for
review.
---
## Where to learn more
- The framework's *why* lives in [the philosophy
document](/philosophy).
- The binding technical contract — section numbers (`§n.n`)
referenced throughout this guide — is in `SPEC.md` in the
framework's source repository.
- Deployment operators have their own recipe in
`docs/DEPLOYMENTS.md`.
+347 -70
View File
@@ -339,6 +339,16 @@ and exact columns are illustrative; the implementing session can adjust.
on first write and updated on every change. Absence of a row means
"no choice yet" — the banner shows. Anonymous viewers persist their
choice in `localStorage` only, with no corresponding row here.
- `device_trust` — per-row record of the §6.2 device-trust gesture
(v0.11.0, roadmap item #9). One row per `(user, trusted device)`
pair; a user with three trusted devices has three rows. Columns:
`id`, `user_id` (FK users, ON DELETE CASCADE), `device_token_hash`
(bcrypt at rest, with a unique index documenting the no-collision
invariant of the 256-bit CSPRNG token space), `created_at`,
`expires_at` (`created_at + 30 days`), `user_agent` (verbatim,
application-layer-truncated to 1024 chars), `last_seen_at`
(refreshed on every successful lookup), `revoked_at` (NULL means
active). The raw token never lives in this table — only the hash.
**Super-draft scoping.** For rows in `threads` and `changes` where the
entry referenced by `rfc_slug` is in state `super-draft`, `branch_name`
@@ -374,7 +384,24 @@ them:
separate "forgot passcode" flow. The user can remove the passcode
at any time from the §6.2 sign-in settings tab, returning to
OTC-only.
3. **Gitea OAuth fallback (migration only).** The v0.1 OAuth
3. **Device trust (cookie-only, 30 days).** Added in v0.11.0
(roadmap item #9). After a successful OTC or passcode sign-in,
the visitor may check "trust this device for 30 days." The
framework then mints a server-issued opaque token, hashes it
(bcrypt) into the `device_trust` table, and sets a long-lived
HttpOnly + Secure + SameSite=Lax cookie carrying the raw token.
On a subsequent visit, `POST /auth/device-trust/start` resolves
the cookie and re-establishes the session without an OTC /
passcode roundtrip. The user can list and revoke their trusted
devices from the `/settings/notifications` "Trusted devices"
section; a revoked or expired cookie is cleared on the next
request. The cookie is "essential" per §14.5 — it is part of
authentication, not analytics, and is set regardless of the
user's analytics / other-cookies choice. The raw token only
ever lives in the outbound `Set-Cookie` header and the inbound
`Cookie` header; server-side storage is the hash, with
constant-time comparison on lookup.
4. **Gitea OAuth fallback (migration only).** The v0.1 OAuth
callback remains functional during the v0.7.0 window, with a
small "Sign in with Gitea (fallback)" link on `/login` so users
with active OAuth sessions or older invite paths still have a
@@ -388,6 +415,34 @@ partial-unique). The Gitea bot user + token are still required for
server-side git operations (repo reads, PR creation); only the
operator-facing sign-in surface moved.
Admission, as of v0.8.0, is by admin grant. v0.7.0 carried the
v0.3.0 `allowed_emails` table forward as the admission gate at the
OTC request surface — emails not on the list got a silent drop.
v0.8.0 (roadmap item #6) reverses that: any valid email receives an
OTC, the fresh `users` row lands in `permission_state='pending'`,
and an admin grant flips the column to `'granted'` before write
endpoints accept the user. The capture-fields step (first name,
last name, free-text "why I should be included in the beta") feeds
the admin's triage queue. The `allowed_emails` table stays in the
schema as a fast-path bypass — the v0.3.0 admin UI still manages
it — but the OTC request path no longer consults it. v0.9.0
(roadmap item #7) shipped the user-management surface that
consumes the `permission_state` column: `/admin/users` carries
every user with their state, profile, and sign-up reason, plus
Grant / Revoke controls that flip the column and write a
`permission_events` audit row. The capture-form submission also
fans a `new_beta_request` notification out to every admin/owner
through the §15 substrate, so the queue surfaces in the inbox +
email channels admins already have.
The `/admin/allowlist` sub-tab stays in place alongside
`/admin/users` rather than merging: the two surfaces have
different keys (allowlist by email pre-sign-up, user list by
user_id post-sign-up) and a union row would be confusing rather
than clarifying. The allowlist's role narrowed to "fast-path
bypass for known-good emails" with the v0.8.0 admission shift;
v0.9.0 retains that role unchanged.
### 6.1 Four roles, each a strict superset of the one below
1. **Anonymous.** Can read public RFCs (the meta repo's main branch,
@@ -404,10 +459,14 @@ operator-facing sign-in surface moved.
surface remain open per the v0.3.0 contract.
2. **Contributor.** Default role for any authenticated account. A
first OTC sign-in by a previously unknown email provisions a row
at this role; v0.7.0 keeps the v0.3.0 allowlist gate (`allowed_emails`)
as the admission control, deferring the open beta-access request
flow to a later release. Everything anonymous can do, plus:
propose new RFCs (open a PR against the meta repo), create
at this role; v0.8.0 replaced the v0.3.0 / v0.7.0 allowlist gate
with an admin-grant flow (roadmap item #6, see opening of §6).
The contributor capabilities below — propose, branch, PR, chat,
claim — are gated by `users.permission_state='granted'` as well
as by the role. A pending contributor (the post-OTC waiting
state) has the same read access as anonymous and zero write
capability until an admin grants. Everything anonymous can do,
plus: propose new RFCs (open a PR against the meta repo), create
branches on any RFC repo, open PRs from branches they have
contribute access to, chat on anything they can read, claim
ownership of unclaimed super-drafts.
@@ -449,6 +508,38 @@ triage what they can't act on, and the restore lands cleanly); a
self-DND'd contributor's own gestures continue to fire signals to
others normally.
v0.8.0 adds a fourth structurally-distinct field on the same row:
`users.permission_state` (the admission gate the v0.8.0 release
ships, see opening of §6 and §6.1). The four — role, muted,
permission_state, the notification mutes — are orthogonal and the
gate semantics compose:
* `role` answers "what scope of action is this user authorized to
perform if they're admitted at all?" (anonymous / contributor /
admin / owner).
* `muted` answers "is this contributor write-restricted by an
admin gesture against their existing grant?" (a sanctions
primitive — owner/admin imposed).
* `permission_state` answers "is this user admitted to the beta
at all?" (the v0.8.0 admin-grant gate — 'pending' / 'granted' /
'revoked'). The default for grandfathered rows at migration time
is `'granted'`; OTC freshly provisions `'pending'`.
* The notification mutes answer "does this user want to receive
signals about a particular RFC or from a particular other
user?" (self-imposed preference).
The four never gate each other. A pending user with `role=owner`
(impossible by construction in v0.8.0 — fresh OTC always provisions
role=contributor — but the orthogonality holds at the column level)
would still refuse write endpoints because the admission gate
runs first; a granted contributor whose row is also muted refuses
writes via the mute gate; a granted contributor with notification
mutes set still passes the contributor gate and writes normally.
v0.6.0's anon-write audit (item #4) is the structural floor for all
four — every write-shaped endpoint funnels through
`auth.require_contributor`, which checks all three of {authenticated,
not muted, permission_state='granted'} in order.
### 6.3 Per-RFC delegated authority
An RFC's `owners:` and `arbiters:` (from the meta-repo entry's
@@ -2033,6 +2124,19 @@ the mechanics, so the mechanics (super-drafts, graduation, public
arguments, AI participation in chat) read as load-bearing rather than
novel.
v0.8.0 (roadmap item #6) added a third sign-in step the surface
runs conditionally — on the first OTC sign-in by a previously
unknown email, the verify response carries `needs_profile=true`
and the surface prompts for first name, last name, and a free-text
"why I should be included in the beta" before bouncing the user to
`/beta-pending`. The page displays a "your request is in review"
message keyed on `users.permission_state='pending'` (repurposed
from the v0.3.0 post-OAuth-rejection surface). Anonymous viewers
and pending viewers see the same read surfaces; only the write
affordances differ. A persistent thin "Your beta access is in
review" banner shows on every page (other than `/beta-pending`
itself) until an admin grants access.
### 14.2 The `/philosophy` route
Authenticated and anonymous visitors alike can reach `/philosophy`,
@@ -2223,8 +2327,14 @@ signal taxonomy this section commits to. The starting set:
`graduation_complete`, `graduation_rolled_back`, `rfc_withdrawn`,
`rfc_reopened`, `claim_opened`, `claim_merged`,
`permission_change_affecting_me`, `app_wide_mute_set`,
`app_wide_mute_lifted`, `digest_emitted`. The enum is extensible; the
build session adjusts as new gestures are wired in.
`app_wide_mute_lifted`, `new_beta_request`, `digest_emitted`.
The enum is extensible; the build session adjusts as new gestures
are wired in. The `new_beta_request` event (v0.9.0, roadmap item
#7) is framework-scoped rather than RFC-scoped — the row's
`rfc_slug` is NULL and the deep-link points `/admin/users`
instead of `/rfc/<slug>` — but otherwise rides the standard
fan-out chokepoint with category `admin-actionable` so the §15.4
email gate only reaches owners/admins.
### 15.2 The inbox
@@ -2669,25 +2779,55 @@ The follow-up session will refine this. A minimal starting set:
- `POST /auth/otc/request` — unauthenticated. Body carries `email`.
Generates a six-digit code, stores its bcrypt hash with an expiry
(`OTC_TTL_MINUTES`, default 10), and dispatches a plain-text email
via the SMTP layer. Returns HTTP 200 (`{ok:true}`) uniformly so
allowlist state (§6.1 / §6.2) is not leaked to callers. Returns
HTTP 429 when the per-email cooldown (`OTC_REQUEST_COOLDOWN_SECONDS`,
default 60) blocks back-to-back requests — the loud-failure shape
for the abuse path. A re-request invalidates the prior unused
code for the same email so only one code is outstanding at a time.
Per §19.2's expected next session, this endpoint is the lead-up
to the Cloudflare-Turnstile abuse-mitigation overlay.
via the SMTP layer. Returns HTTP 200 (`{ok:true}`) uniformly.
Returns HTTP 429 when the per-email cooldown
(`OTC_REQUEST_COOLDOWN_SECONDS`, default 60) blocks back-to-back
requests — the loud-failure shape for the abuse path. A re-request
invalidates the prior unused code for the same email so only one
code is outstanding at a time. v0.7.0 also dropped requests
silently if the email wasn't on the `allowed_emails` list (the
v0.3.0 admission gate); v0.8.0 (item #6) removed that check —
admission moved to `permission_state` on the freshly-provisioned
`users` row, asserted at the contributor gate. v0.12.0 (item #10)
gates this endpoint behind a CloudFlare Turnstile siteverify call:
the body carries an optional `turnstile_token` field, the server
POSTs `secret` + `response` to `challenges.cloudflare.com/turnstile/
v0/siteverify` before the bcrypt hash + SMTP send, and a failed
challenge refuses with HTTP 400 spending no rate budget. Two env
vars drive the policy: `CLOUDFLARE_TURNSTILE_SECRET` (Secret
Manager) and `TURNSTILE_REQUIRED` (overlay, default `false`). When
the secret is unset and `TURNSTILE_REQUIRED=false`, the gate is
open (the dev / pre-rollout path); when the secret is unset and
`TURNSTILE_REQUIRED=true`, the endpoint refuses with HTTP 500
"auth misconfigured" so a future config drift fails loudly
instead of silently disabling abuse defense.
- `POST /auth/otc/verify` — unauthenticated. Body carries `email` and
`code`. Validates the bcrypt hash against the most-recent unconsumed
non-expired row for the email, marks the row consumed, provisions
or links the `users` row by email (per §6.2's migration path —
match by `users.email` case-insensitive, otherwise insert a fresh
contributor row with `gitea_id = NULL`), and stores the session
cookie. Returns HTTP 200 on success with a minimal user payload;
HTTP 400 on any failure (expired, consumed, wrong, unknown). The
failure modes collapse to a single generic message so a probing
client cannot distinguish "you got the wrong code" from "we don't
know this email" — the operator logs carry the distinction.
contributor row with `gitea_id = NULL` and
`permission_state='pending'`), and stores the session cookie.
Returns HTTP 200 on success; the response body carries
`{ok, user, needs_profile}` where `needs_profile=true` iff the
user is `permission_state='pending'` AND the row has no
first_name / last_name / beta_request_reason yet (a fresh OTC
sign-in). The `needs_profile` flag drives the Login.jsx surface's
step-3 capture form. HTTP 400 on any failure (expired, consumed,
wrong, unknown). The failure modes collapse to a single generic
message so a probing client cannot distinguish "you got the
wrong code" from "we don't know this email" — the operator logs
carry the distinction.
- `POST /api/auth/me/beta-request` — authenticated. Body carries
`first_name`, `last_name`, `beta_request_reason` (all required;
bounded at 120 / 120 / 4000 chars). Writes the fields to the
signed-in user's row and leaves `permission_state='pending'`.
Idempotent for the same already-pending user (a re-submit
updates the row so the admin sees the latest text). Refuses
with HTTP 409 if the user is already `'granted'` or `'revoked'`.
v0.8.0 — the first-OTC profile-capture endpoint (roadmap item
#6). v0.9.0's admin user-management page consumes this column
set to render the request queue.
- `GET /auth/passcode/check` — unauthenticated. Query param `email`.
Returns `{has_passcode: boolean}`. The frontend's `/login` surface
calls this after the email step to decide whether to render a
@@ -2720,6 +2860,28 @@ The follow-up session will refine this. A minimal starting set:
return HTTP 400 with a generic message; the no-passcode-set
failure also collapses to 400 so the response does not enumerate
account state. v0.10.0.
- `POST /auth/device-trust/start` — unauthenticated. Reads the
`rfc_device_trust` cookie (set previously by an OTC or passcode
verify with `trust_device: true`). On a non-expired, non-revoked
match, re-establishes the session and returns HTTP 200 with the
minimal user payload. On a miss (no cookie, expired, revoked, or
unknown), returns HTTP 401 and clears the stale cookie via the
response's `Set-Cookie` header. The failure modes collapse to
one shape so a probing client cannot enumerate "your row was
revoked" vs. "this token never existed". v0.11.0.
- `GET /api/auth/me/devices` — authenticated. Returns the active
(`revoked_at IS NULL` AND `expires_at > now`) device-trust rows
for the signed-in user: `id`, `created_at`, `expires_at`,
`last_seen_at`, `user_agent`. The bcrypt hash is structurally
private and is never surfaced. v0.11.0.
- `DELETE /api/auth/me/devices/{id}` — authenticated. Stamps
`revoked_at` on the row with id `{id}` belonging to the
signed-in user. The user-id scope is enforced in SQL so a
hostile client cannot revoke another user's row by guessing
ids; a row that does not match returns HTTP 404. v0.11.0.
- `DELETE /api/auth/me/devices` — authenticated. Revokes every
active row for the signed-in user; returns the count revoked.
v0.11.0.
- `GET /api/rfcs` — list entries with state, id, title, slug, repo,
owners, last_active_at, has_open_prs, starred-by-me. Supports
search, sort, filter chips, and the `unclaimed` predicate.
@@ -2862,8 +3024,15 @@ The follow-up session will refine this. A minimal starting set:
- `POST /api/rfcs/<slug>/prs/<pr_number>/withdraw` — withdraw per §10.8.
- `POST /api/rfcs/<slug>/prs/<pr_number>/resolution-branch` — cut a
fresh resolution branch and replay per §10.9.
- `GET /api/admin/users` — list users with role and write-mute state,
for the §6 / Slice 7 admin surface.
- `GET /api/admin/users` — list users for the §6 / Slice 7 admin
surface. v0.9.0 (roadmap item #7) widened the payload to carry
`permission_state`, `first_name`, `last_name`, `beta_request_reason`,
`created_at`, `permission_decided_at`, and the joined
`permission_decided_by_login` / `permission_decided_by_display`
for the user-management page. Sort order surfaces `pending` rows
first (the daily admin queue), then `granted`, then `revoked`;
within a bucket, owners precede admins precede contributors,
with recency as the tiebreaker.
- `POST /api/admin/users/<id>/role` — set role. Only owners may grant
or revoke `owner`; admins may flip contributor ↔ admin freely. An
owner-self-demotion is refused on this endpoint; owner succession
@@ -2872,6 +3041,16 @@ The follow-up session will refine this. A minimal starting set:
write-mute (not the §15.8 notification mutes). Refused on owners
and admins — for them, the role-change channel is the right
refusal. Writes a `permission_events` row.
- `POST /api/admin/users/<id>/permission` — v0.9.0 (roadmap item #7).
Flip `permission_state` between `pending`, `granted`, and `revoked`.
Stamps `permission_decided_by` + `permission_decided_at` on the
row and writes a `permission_events` row with event_kind in
`{permission_granted, permission_revoked, permission_repended}`.
Refuses with 422 if the admin tries to flip their own row
(symmetric to the `set_mute` / `set_role` self-action refusals).
v0.8.0 shipped the column shape with no admin UI — operators ran
a manual `UPDATE users` to grant access; v0.9.0 retires the
manual gesture.
- `GET /api/admin/audit` — paged read of the `actions` log with
filters `action_kind`, `actor_user_id`, `rfc_slug`, plus `before_id`
for the page boundary. Returns the joined actor login/display so
@@ -3695,22 +3874,80 @@ the new §15 (Notifications, in full), and §17 (the notification
endpoints — list, mark-read, stream, watch mutation, preferences,
quiet-hours, per-user mute, unsubscribe, bounce webhook).
- **First-OTC profile capture.** *Surfaced by v0.7.0's email/OTC
migration.* When a fresh email lands at `/auth/otc/verify` with
no matching `users.email` row, v0.7.0 provisions the row with
`display_name = <local part of email>` and no other identity
fields. The roadmap item-#6 candidate (v0.8.0) is expected to
add a one-shot profile-capture step on the first-OTC sign-in:
first name, last name, and a free-text "why I want access" field
that flows into the open beta-access request queue (also item #6)
that replaces the v0.3.0 `allowed_emails` gate. The schema slot
exists implicitly already (`users.display_name` is updateable,
the audit-log + permission-events tables carry the freeform
notes); the structural decision is what gates the capture (modal
on `/login` after verify? a one-time redirect to `/welcome/profile`?
a deferred banner on the main view?) and how it interacts with
the open-access request flow that replaces the allowlist. Earns
its session as the v0.8.0 design pass.
First-OTC profile capture (formerly a v0.7.0 candidate) is settled
and folded into §6.1 (the contributor role now requires
`permission_state='granted'`), §6.2 (the orthogonality of
permission_state vs role / muted / notification-mutes), §14.1
(the landing page's v0.8.0 first-OTC capture step), and §17
(the `POST /api/auth/me/beta-request` endpoint and the verify
endpoint's new `needs_profile` flag). The structural decision
landed as: capture is a third step on the `/login` surface
gated by `verify_response.needs_profile=true`; pending users
land on `/beta-pending` after submitting and see a thin banner
on every other page until an admin grants. v0.8.0 (roadmap item
#6) shipped the work.
Candidates surfaced during v0.8.0 (open beta-access request flow,
§6.1 / §14.1, item #6):
- **Admin user-management page** (`/admin/users`). *Shipped in
v0.9.0 (roadmap item #7).* The listing surfaces every user with
permission_state, profile fields, sign-up reason, and a Grant /
Revoke control set; the `POST /api/admin/users/<id>/permission`
endpoint flips the column and writes a `permission_events` row.
v0.9.0 left the `/admin/allowlist` sub-tab in place rather than
merging (see allowlist deprecation below). The grant/revoke
notify-the-user surface is deferred (see the new candidate
below).
- **Allowlist deprecation.** *Decision deferred past v0.9.0.*
v0.9.0 considered merging `/admin/allowlist` into the new
`/admin/users` page but kept the surface as a sibling sub-tab:
the two have different keys (allowlist by email pre-sign-up,
user list by user_id post-sign-up) and a union row would be
confusing rather than clarifying. The fast-path-bypass role
the allowlist has carried since v0.8.0 stays intact; the
cutover to retire the table outright is a later session.
Decision points unchanged from v0.8.0: drop the table outright
(a schema migration) or leave it as a non-functional surface
and remove only the UI (a frontend-only change); how to handle
existing `allowed_emails` rows at the cutover (probably: walk
them into the pending queue with `permission_state='granted'`
for any matching `users` row, leave unmatched rows as a no-op
since v0.8.0 doesn't consult them anymore). Earns its session
once the v0.9.0 admin queue has run long enough to confirm the
allowlist's bypass role is no longer pulling weight.
- **Admin notification on new beta request.** *Shipped in v0.9.0
(roadmap item #7).* The `POST /api/auth/me/beta-request`
handler now calls `notify.fan_out_new_beta_request`, which
fans a `new_beta_request` event (category `admin-actionable`,
rfc_slug NULL) out to every owner / admin. The §15 chokepoint
handles the SSE broadcast and the §15.4 email dispatch; the
email reaches only recipients whose `email_admin_actionable`
toggle is on (the default for owners + admins).
- **Grant / revoke notification to the user.** *Surfaced by
v0.9.0.* The new flip endpoint stamps `permission_decided_by` +
writes a `permission_events` row but does not yet signal the
affected user that their state changed. A future release could
fire a `personal-direct` notification (event_kind
`permission_change_affecting_me`, already in the §15.1 enum) so
a granted user sees "Your beta-access request was approved" in
their inbox and email, and a revoked user sees a parallel
refusal notice. Decision points: does revocation include a
reason field (probably yes — symmetric with §9.3's decline
comment); does grant carry a welcome message (probably no —
the existing welcome surfaces are sufficient); does the
notification escape the §15.8 mute path (probably yes — it's
a personal-direct admission state change). Earns its session
as a follow-up to the v0.9.0 page.
- **Decline-with-reason on permission revoke.** *Surfaced by
v0.9.0.* The current Revoke gesture takes only a confirmation;
there is no audit-visible reason captured. A future release
could add a free-text reason input that lands in the
`permission_events.details` JSON column (no schema change
needed — the column is already JSON-shaped). This is the
symmetric companion to the §9.3 proposal-decline contract.
Earns its session alongside the grant/revoke notification
candidate above.
- **Removing the Gitea OAuth fallback.** *Surfaced by v0.7.0.*
v0.7.0 keeps `/auth/callback` functional and links to it as a
"Sign in with Gitea (fallback)" affordance on the new `/login`
@@ -3725,37 +3962,77 @@ quiet-hours, per-user mute, unsubscribe, bounce webhook).
message), and whether the `/auth/login` and `/auth/callback`
routes get a tombstone redirect to `/login` or just 404. Earns
its session once the OTC adoption curve flattens.
- **Device trust (30-day skip).** *Surfaced by v0.7.0 — the
signed-in cookie already lasts 30 days via SessionMiddleware,
but every sign-in still requires a fresh OTC or passcode.* The
roadmap item-#9 candidate adds a "trust this device" affordance
on the verify step that issues a longer-lived rotating token,
so returning visitors on the same device skip both the OTC and
the passcode step. The shape question is whether the trust is a
signed cookie distinct from the session, a row in a `device_trust`
table keyed by a random device-id, or a property of the session
itself; and whether the trust survives password-equivalent events
— v0.10.0's passcode-change and passcode-clear gestures are the
v1 instances — or only survives explicit logout. Earns its
session as the v0.11.0 design pass.
- **Cloudflare Turnstile (or equivalent) on `/auth/otc/request`.**
*Surfaced by v0.7.0 — the endpoint is now the new abuse hot
path.* Per-email cooldown stops the trivial loop; what it
doesn't stop is a distributed scrape that fans out across a
large invitee list to harvest the "this email is admitted vs.
this email is not" signal indirectly (timing differences, SMTP
bounce-rate observation). The roadmap item-#10 candidate gates
the request endpoint behind a one-step browser-side challenge
before the bcrypt hash + SMTP send. Open questions: which
provider (Turnstile is the default since it's free and
privacy-respecting; hCaptcha and reCAPTCHA are also viable);
how the deployment configures it (`TURNSTILE_SITE_KEY` +
`TURNSTILE_SECRET_KEY` env vars, gated by `if
config.turnstile_site_key:` at the handler so existing
deployments don't break); whether the verify endpoint also
gets a challenge (probably yes for parity); and how the test
harness mocks the challenge. Earns its session as the v0.12.0
design pass.
- **Device trust (30-day skip).** *Settled in v0.11.0 (roadmap
item #9). The shape: a distinct `rfc_device_trust` cookie
(HttpOnly + Secure + SameSite=Lax + 30-day Max-Age) carrying a
server-issued opaque token, keyed against a `device_trust` table
whose rows store the bcrypt hash. `POST /auth/device-trust/start`
resolves a presented cookie at next visit. A
`/settings/notifications` "Trusted devices" section lists active
rows with per-row + bulk revoke. The trust outlives a sign-out
(sign-out clears the session cookie, not the device-trust
cookie) and is not affected by passcode set/change/clear — the
next two items below carry the remaining open questions.*
- **Cross-device session revocation surface.** v0.11.0's
`/settings/notifications → Trusted devices` revokes the
long-lived device-trust grants. What it does NOT revoke is an
active session cookie sitting in another browser, or the
v0.10.0 passcode-failure-counter shape, or a stale
password-equivalent that some future release ships. The natural
next step is a single "active sessions and devices" surface
that lists everything currently authenticating as this user —
device-trust rows + active session cookies (if/when the
framework moves to server-side sessions) + future credential
shapes — and lets the user kill any of them with one gesture.
Earns its session when a second cross-cutting concern lands
(the most likely first trigger: future Yubikey / WebAuthn
support, which surfaces another credential to revoke).
- **Password-equivalent change invalidates device trust.** v0.11.0
intentionally leaves device-trust rows live across a passcode
set / change / clear. The argument is structural: the user has
the v0.10.0 lockout, the v0.11.0 per-device revoke list, and a
fresh sign-in path via OTC, so the cookie is not a high-value
bypass relative to the keys-to-the-account a passcode change
signals. The argument against is the conventional "changing a
password should kill every active session" expectation users
bring from other systems. This earns its own session once the
evidence is in: either a security-review finding that says
"this is the wrong default," or user feedback that says "I
expected my old laptop to sign out when I changed my passcode."
- **Device-trust window tunables via env.** v0.11.0 hard-codes
the 30-day window in `backend/app/device_trust.py`
(`TRUST_DURATION_DAYS = 30`). Surfacing it as an env var
(`DEVICE_TRUST_DURATION_DAYS`?) is small and obvious; deferring
follows the same pattern as the v0.10.0 passcode-lockout
hard-coding — name the tunable when a deployment wants it
different rather than shipping a knob that has no operator
asking for it.
- **Cloudflare Turnstile on `/auth/otc/request`.** *Settled in
v0.12.0 (roadmap item #10).* The OTC request endpoint is now
gated behind a Turnstile siteverify call: the frontend renders
the official widget on the `/login` email-entry step (and on the
passcode step for the "Use a code instead" fallback dispatch),
the captured token rides in the request body as
`turnstile_token`, and the backend POSTs `secret` + `response`
to `challenges.cloudflare.com/turnstile/v0/siteverify` before
the bcrypt hash + SMTP send. The widget renders only on the
email-entry / passcode-fallback dispatch points — the OTC
verify step is bottlenecked on email delivery and protected by
the five-minute TTL + single-use row consume, so a second
challenge there would double the rate budget against the same
abuse path without measurably more protection (revisit if bots
adapt to the email-entry challenge specifically). Configured
via `CLOUDFLARE_TURNSTILE_SECRET` (Secret Manager) and
`VITE_TURNSTILE_SITE_KEY` (frontend build-time overlay); a
third var `TURNSTILE_REQUIRED` (default `false`) lets the
operator flip from "soft-fail when secret unset" (the dev /
pre-rollout shape) to "fail-closed when secret unset" (HTTP 500
"auth misconfigured", the production-locked shape). Tests mock
the siteverify HTTP call at the `httpx.post` boundary in
`app.turnstile`. The hCaptcha / reCAPTCHA alternatives noted in
the v0.7.0 surfacing are still viable substitutes for a future
deployment that wants them but the framework's tested path is
Turnstile.
Candidates surfaced during v0.10.0 (user-set passcodes, §6.2 /
roadmap item #8):
+1 -1
View File
@@ -1 +1 @@
0.10.0
0.12.0
+18
View File
@@ -92,3 +92,21 @@ OTC_TTL_MINUTES=10
# loud-failure shape so the abuse path is visible). Set to 0 to
# disable the cooldown — useful for tests but never in production.
OTC_REQUEST_COOLDOWN_SECONDS=60
# --- v0.12.0: CloudFlare Turnstile gate on OTC dispatch (§6.2, item #10) ---
# Provision a Turnstile site at dash.cloudflare.com → Turnstile → Add
# site. The site key (public) goes in `frontend/.env` as
# VITE_TURNSTILE_SITE_KEY. The secret key (private) goes here and is
# what the backend POSTs to /siteverify alongside the user's response
# token. Leave both unset for dev/test paths; the gate stays open when
# the secret is absent AND TURNSTILE_REQUIRED=false (the default).
CLOUDFLARE_TURNSTILE_SECRET=
# When `true`, /auth/otc/request fails closed (HTTP 500 "auth
# misconfigured") if CLOUDFLARE_TURNSTILE_SECRET is unset. When `false`
# (the default), a missing secret skips verification — useful in dev
# and during the pre-rollout window when the operator hasn't wired
# the secret yet. Flip to `true` once the secret is wired so a future
# config drift surfaces as a loud 500 rather than a silent abuse-
# defense disablement.
TURNSTILE_REQUIRED=false
+176 -6
View File
@@ -26,10 +26,13 @@ from . import (
api_prs,
auth,
db,
device_trust as device_trust_mod,
docs as docs_mod,
entry as entry_mod,
cache,
funder,
health,
notify,
philosophy,
providers as providers_mod,
)
@@ -55,6 +58,17 @@ class FunderCredentialBody(BaseModel):
api_key: str = Field(min_length=1, max_length=2048)
class BetaRequestBody(BaseModel):
# v0.8.0 — captured on the first OTC sign-in. All three fields are
# required so the admin queue has a coherent triage shape.
# The bounds match the v0.7.0 OTC body (320 chars for email-ish
# headers; 4000 for the free-text reason — the same upper bound
# DeclineBody uses elsewhere in this file).
first_name: str = Field(min_length=1, max_length=120)
last_name: str = Field(min_length=1, max_length=120)
beta_request_reason: str = Field(min_length=1, max_length=4000)
def make_router(
config: Config,
gitea: Gitea,
@@ -111,6 +125,17 @@ def make_router(
payload = philosophy.load()
return {"body": payload["body"]}
# ---------------------------------------------------------------
# /api/docs — DOCS.md served verbatim. Sibling of /api/philosophy:
# no auth gate, same disk-first load + cache shape, same intent —
# public read surface for a markdown file checked into the repo.
# ---------------------------------------------------------------
@router.get("/api/docs")
async def get_docs() -> dict[str, Any]:
payload = docs_mod.load()
return {"body": payload["body"]}
# ---------------------------------------------------------------
# Auth surface — reads role from our users table per §6.
# ---------------------------------------------------------------
@@ -120,15 +145,27 @@ def make_router(
user = auth.current_user(request)
if user is None:
return {"authenticated": False, "user": None}
# v0.10.0: surface a single `has_passcode` flag so the §6.2
# settings tab can render "Set passcode" vs. "Change/Remove
# passcode" without a second round trip. The set-at timestamp
# rides along for the same reason. The hash itself is never
# exposed.
# v0.8.0 + v0.10.0: single round-trip for everything the
# frontend gates UI off of — beta-access state + passcode state.
row = db.conn().execute(
"SELECT passcode_hash, passcode_set_at FROM users WHERE id = ?",
"SELECT first_name, last_name, beta_request_reason, "
"passcode_hash, passcode_set_at "
"FROM users WHERE id = ?",
(user.user_id,),
).fetchone()
first_name = (row["first_name"] if row else None) or ""
last_name = (row["last_name"] if row else None) or ""
beta_request_reason = (row["beta_request_reason"] if row else None) or ""
# "Needs profile" iff the user is pending AND hasn't yet
# filed their beta-request capture. Granted users never see
# the capture prompt; pending users who already filed see
# the /beta-pending page without the capture form.
needs_profile = (
user.permission_state == "pending"
and not first_name
and not last_name
and not beta_request_reason
)
has_passcode = bool(row and row["passcode_hash"])
passcode_set_at = row["passcode_set_at"] if (row and has_passcode) else None
return {
@@ -140,11 +177,144 @@ def make_router(
"email": user.email,
"avatar_url": user.avatar_url,
"role": user.role,
"permission_state": user.permission_state,
"first_name": first_name,
"last_name": last_name,
"beta_request_reason": beta_request_reason,
"needs_profile": needs_profile,
"has_passcode": has_passcode,
"passcode_set_at": passcode_set_at,
},
}
# ---------------------------------------------------------------
# v0.8.0: /api/auth/me/beta-request — first-OTC profile capture
# (roadmap item #6). Lands first name, last name, and the free-
# text "why I should be included in the beta" on the signed-in
# user's row. Idempotent for the same already-pending user;
# refuses to overwrite a row that's already granted (so a
# bored already-granted user can't accidentally re-submit the
# form and clobber the admin's audit trail). Uses
# `require_user` rather than `require_contributor` because
# `require_contributor` already enforces `permission_state =
# 'granted'` and would refuse a pending user; the whole point
# of this endpoint is to register the request _from_ a pending
# user.
# ---------------------------------------------------------------
@router.post("/api/auth/me/beta-request")
async def submit_beta_request(body: BetaRequestBody, request: Request) -> dict[str, Any]:
user = auth.require_user(request)
row = db.conn().execute(
"SELECT permission_state, first_name, last_name, beta_request_reason FROM users WHERE id = ?",
(user.user_id,),
).fetchone()
if row is None:
# Defensive — the session pointed at a deleted row.
raise HTTPException(404, "User not found")
# Granted users have no business filing a beta request.
# 'revoked' likewise — the request flow is for fresh users
# only. Both shapes refuse with 409 (conflict) so the client
# can distinguish "you already have access" from
# "your access was revoked".
if row["permission_state"] == "granted":
raise HTTPException(409, "Your account is already granted access")
if row["permission_state"] == "revoked":
raise HTTPException(409, "Your account's access has been revoked")
# Re-submission from a pending user updates the row — the
# admin sees the latest text rather than a stale draft.
# The state stays 'pending'; only an admin can flip it.
db.conn().execute(
"""
UPDATE users
SET first_name = ?,
last_name = ?,
beta_request_reason = ?
WHERE id = ?
""",
(
body.first_name.strip(),
body.last_name.strip(),
body.beta_request_reason.strip(),
user.user_id,
),
)
# v0.9.0 (roadmap item #7): notify every admin/owner of the
# fresh request. Only the first submission is the
# "newly-pending" gesture — re-submits from the same user
# would otherwise carpet the admin inbox. We fire only when
# this is the row's first time getting all three fields
# populated (the prior row carried at least one NULL).
prior = row # captured before the UPDATE above
was_already_complete = bool(
prior["first_name"] and prior["last_name"] and prior["beta_request_reason"]
)
if not was_already_complete:
notify.fan_out_new_beta_request(requester_user_id=user.user_id)
return {"ok": True}
# ---------------------------------------------------------------
# v0.11.0: trust device for 30 days (§6.2, roadmap item #9).
#
# The mint path lives on the OAuth router (issuing the cookie is
# coupled to OTC/passcode verify). This module owns the read/revoke
# surface the /settings/devices page calls.
# ---------------------------------------------------------------
@router.get("/api/auth/me/devices")
async def list_my_devices(request: Request) -> dict[str, Any]:
"""Active device-trust rows for the signed-in user.
Active = not revoked, not expired. The current request's
device (if any) is *not* singled out here the surface
shows the same row shape for every device so the user can
revoke any of them without the page leaking which row
carries the cookie they're using right now.
"""
user = auth.require_user(request)
rows = device_trust_mod.list_for_user(user.user_id)
return {
"items": [
{
"id": r.id,
"created_at": r.created_at,
"expires_at": r.expires_at,
"last_seen_at": r.last_seen_at,
"user_agent": r.user_agent,
}
for r in rows
]
}
@router.delete("/api/auth/me/devices/{device_id}")
async def revoke_my_device(device_id: int, request: Request) -> dict[str, Any]:
"""Revoke a single device-trust row for the signed-in user.
The user-id scope is enforced in SQL so a hostile client
cannot revoke another user's row by guessing ids. A row that
doesn't exist, doesn't belong to this user, or is already
revoked reads as 404 the wrong-vs-already-revoked
distinction would only help a probing client enumerate ids.
"""
user = auth.require_user(request)
ok = device_trust_mod.revoke(user.user_id, device_id)
if not ok:
raise HTTPException(404, "Device not found")
return {"ok": True}
@router.delete("/api/auth/me/devices")
async def revoke_all_my_devices(request: Request) -> dict[str, Any]:
"""Revoke every active device-trust row for the signed-in user.
The user's current request stays authenticated via its
session cookie; the device-trust cookie carried on the
current device is also revoked, but `rfc_session` keeps the
request flow alive until sign-out / expiry.
"""
user = auth.require_user(request)
count = device_trust_mod.revoke_all(user.user_id)
return {"ok": True, "revoked": count}
# ---------------------------------------------------------------
# §7: the catalog
# ---------------------------------------------------------------
+117 -4
View File
@@ -50,6 +50,16 @@ class MuteBody(BaseModel):
muted: bool
class PermissionStateBody(BaseModel):
# v0.9.0: the admin flip from the user-management page (roadmap
# item #7). `pending` is not surfaceable from the admin UI —
# only the OTC verify path lands a row in `pending` — but we
# accept it in the pattern in case a future restore-to-queue
# gesture wants to re-pend a granted user; today the UI only
# exposes `granted` and `revoked`.
state: str = Field(pattern="^(pending|granted|revoked)$")
class AllowlistAddBody(BaseModel):
email: str = Field(min_length=3, max_length=320)
note: str | None = Field(default=None, max_length=200)
@@ -68,13 +78,41 @@ def make_router(config: Config) -> APIRouter:
@router.get("/api/admin/users")
async def list_users(request: Request) -> dict[str, Any]:
"""v0.9.0: the user-management surface (roadmap item #7).
The listing carries every column the admin queue needs to triage
pending beta-access requests alongside the existing role/mute
affordances. Sort order surfaces pending requests first (so the
admin lands on the inbox shape), then granted, then revoked;
within a state, ownership/role and recency are the tiebreakers
so the legacy ordering (owner first, then admin, then by name)
is preserved inside the granted bucket.
`permission_decided_by_login` joins the deciding admin row so
the UI can render "granted by @ben" without a second round-trip.
"""
auth.require_admin(request)
rows = db.conn().execute(
"""
SELECT id, gitea_login, display_name, email, role, muted,
created_at, last_seen_at
FROM users
ORDER BY role = 'owner' DESC, role = 'admin' DESC, display_name COLLATE NOCASE
SELECT u.id, u.gitea_login, u.display_name, u.email, u.role, u.muted,
u.created_at, u.last_seen_at,
u.permission_state, u.first_name, u.last_name,
u.beta_request_reason,
u.permission_decided_by, u.permission_decided_at,
d.gitea_login AS decided_by_login,
d.display_name AS decided_by_display
FROM users u
LEFT JOIN users d ON d.id = u.permission_decided_by
ORDER BY
CASE u.permission_state
WHEN 'pending' THEN 0
WHEN 'granted' THEN 1
WHEN 'revoked' THEN 2
ELSE 3
END,
u.role = 'owner' DESC, u.role = 'admin' DESC,
COALESCE(u.last_seen_at, u.created_at) DESC,
u.display_name COLLATE NOCASE
"""
).fetchall()
return {
@@ -88,6 +126,13 @@ def make_router(config: Config) -> APIRouter:
"muted": bool(r["muted"]),
"created_at": r["created_at"],
"last_seen_at": r["last_seen_at"],
"permission_state": r["permission_state"] or "granted",
"first_name": r["first_name"] or "",
"last_name": r["last_name"] or "",
"beta_request_reason": r["beta_request_reason"] or "",
"permission_decided_at": r["permission_decided_at"],
"permission_decided_by_login": r["decided_by_login"],
"permission_decided_by_display": r["decided_by_display"],
}
for r in rows
]
@@ -136,6 +181,74 @@ def make_router(config: Config) -> APIRouter:
)
return {"ok": True, "role": body.role, "changed": True}
# ----- Permission state (§6.1, v0.9.0 roadmap item #7) -----
@router.post("/api/admin/users/{user_id}/permission")
async def set_permission(user_id: int, body: PermissionStateBody, request: Request) -> dict[str, Any]:
"""Flip a user's `permission_state` between pending/granted/revoked.
v0.8.0 wired the column shape but shipped no admin UI for it
the grant gesture was a manual `UPDATE users` against the DB.
v0.9.0 (roadmap item #7) lands the admin user-management page;
this endpoint is its single write surface.
Audit shape: every flip writes a `permission_events` row with
event_kind in {'permission_granted', 'permission_revoked',
'permission_repended'} so §6.5's log carries the change. The
`permission_decided_by` / `permission_decided_at` columns on
the user row are co-stamped so the user listing can render
"granted by @ben at <date>" without a second join through
the audit table.
Refuses with 422 if the admin tries to flip their own row
(no self-grant / self-revoke; symmetric to set_mute's
self-mute refusal and set_role's self-downgrade refusal).
"""
viewer = auth.require_admin(request)
target = db.conn().execute(
"SELECT id, role, permission_state FROM users WHERE id = ?",
(user_id,),
).fetchone()
if target is None:
raise HTTPException(404, "User not found")
if target["id"] == viewer.user_id:
raise HTTPException(422, "You cannot change your own permission state")
before = target["permission_state"] or "granted"
after = body.state
if before == after:
return {"ok": True, "permission_state": after, "changed": False}
db.conn().execute(
"""
UPDATE users
SET permission_state = ?,
permission_decided_by = ?,
permission_decided_at = datetime('now')
WHERE id = ?
""",
(after, viewer.user_id, user_id),
)
event_kind = {
"granted": "permission_granted",
"revoked": "permission_revoked",
"pending": "permission_repended",
}[after]
db.conn().execute(
"""
INSERT INTO permission_events
(actor_user_id, subject_user_id, event_kind, details)
VALUES (?, ?, ?, ?)
""",
(
viewer.user_id,
user_id,
event_kind,
json.dumps({"before": before, "after": after}),
),
)
return {"ok": True, "permission_state": after, "changed": True}
# ----- Write-mute (§6.2) -----
@router.post("/api/admin/users/{user_id}/mute")
+60 -4
View File
@@ -30,6 +30,12 @@ class SessionUser:
email: str
avatar_url: str
role: str
# v0.8.0 / §6.1 — admission gate. Three states: 'pending' (waiting
# for an admin grant), 'granted' (active contributor), 'revoked'
# (was granted, later removed). Existing rows at migration time
# default to 'granted' so grandfathered users are unaffected; OTC
# provisions fresh users with 'pending' (see `app/otc.py`).
permission_state: str = "granted"
def as_actor(self) -> Actor:
return Actor(
@@ -90,6 +96,13 @@ def allowlist_is_active() -> bool:
def is_allowed_sign_in(profile: dict[str, Any]) -> bool:
"""Decide whether a freshly-completed OAuth profile may sign in.
v0.8.0 (item #6) replaces the allowlist gate with an admin-grant
flow at the OTC `/request` surface, but the Gitea OAuth callback
in `main.py` still consults this helper so the fallback path
keeps the v0.3.0 admission shape during the OAuth migration
window. The eventual removal of the OAuth callback (§19.2)
retires this function alongside it.
Three accept paths:
1. The allowlist is empty (gate off).
2. The Gitea profile's email is in `allowed_emails` (case-insensitive).
@@ -132,17 +145,27 @@ def provision_user(config: Config, profile: dict[str, Any]) -> SessionUser:
existing = c.execute("SELECT * FROM users WHERE gitea_id = ?", (gitea_id,)).fetchone()
if existing is None:
role = "owner" if config.owner_gitea_login and login == config.owner_gitea_login else "contributor"
# v0.8.0: a fresh OAuth-provisioned user is also subject to
# the admin-grant flow. The OAuth fallback only fires for
# users who pass `is_allowed_sign_in` (so they're already on
# the legacy allowlist or are grandfathered by gitea_id);
# 'granted' is the right default here since the allowlist
# check is itself the admin gesture. A future release that
# retires the OAuth callback (§19.2) collapses both paths
# under the same gate.
cur = c.execute(
"""
INSERT INTO users (gitea_id, gitea_login, email, display_name, avatar_url, role)
VALUES (?, ?, ?, ?, ?, ?)
INSERT INTO users (gitea_id, gitea_login, email, display_name, avatar_url, role, permission_state)
VALUES (?, ?, ?, ?, ?, ?, 'granted')
""",
(gitea_id, login, email, display, avatar, role),
)
user_id = cur.lastrowid
permission_state = "granted"
else:
user_id = existing["id"]
role = existing["role"]
permission_state = existing["permission_state"] or "granted"
c.execute(
"""
UPDATE users
@@ -160,6 +183,7 @@ def provision_user(config: Config, profile: dict[str, Any]) -> SessionUser:
email=email,
avatar_url=avatar,
role=role,
permission_state=permission_state,
)
@@ -178,6 +202,12 @@ def store_session(request: Request, user: SessionUser) -> None:
"email": user.email,
"avatar_url": user.avatar_url,
"role": user.role,
# v0.8.0: persist the admission state on the cookie payload so
# the post-cookie audit doesn't second-guess the row. The DB
# is re-read on every `current_user` call regardless (so an
# admin grant takes effect on the next request); this field
# is purely structural redundancy for the cookie shape.
"permission_state": user.permission_state,
}
@@ -188,7 +218,7 @@ def current_user(request: Request) -> SessionUser | None:
# Re-read the role from the database every request so role changes
# take effect on the next API call without forcing a logout.
row = db.conn().execute(
"SELECT id, gitea_id, gitea_login, email, display_name, avatar_url, role FROM users WHERE id = ?",
"SELECT id, gitea_id, gitea_login, email, display_name, avatar_url, role, permission_state FROM users WHERE id = ?",
(raw["user_id"],),
).fetchone()
if row is None:
@@ -199,6 +229,11 @@ def current_user(request: Request) -> SessionUser | None:
# of which sign-in path the row came from. The DB remains the
# source of truth for "is this an OAuth-linked user" (gitea_id IS
# NOT NULL); the in-memory SessionUser is the per-request handle.
# v0.8.0: permission_state comes off the row directly. A NULL
# column value (shouldn't happen under the migration's
# NOT NULL DEFAULT, but be defensive) reads as 'granted' so the
# gate fails open for grandfathered surfaces rather than locking
# everyone out on a malformed row.
return SessionUser(
user_id=row["id"],
gitea_id=row["gitea_id"] or 0,
@@ -207,6 +242,7 @@ def current_user(request: Request) -> SessionUser | None:
email=row["email"] or "",
avatar_url=row["avatar_url"] or "",
role=row["role"],
permission_state=row["permission_state"] or "granted",
)
@@ -218,11 +254,31 @@ def require_user(request: Request) -> SessionUser:
def require_contributor(request: Request) -> SessionUser:
"""§6.1: authenticated, not write-muted."""
"""§6.1: authenticated, not write-muted, and granted by an admin.
v0.8.0 (item #6) widens this gate. A fresh OTC sign-in lands in
`permission_state='pending'`; the user can read everything an
anonymous viewer can read, but every write-shaped endpoint that
funnels through this dependency now refuses with 403 until an
admin grants them. The `pending` blast radius is the same as
anonymous (item #4 / v0.6.0 already audited the anon-write
refusal at every write site), so this widening is structurally
a relabel the same surfaces that already refused 401 to
anonymous now also refuse 403 to pending.
"""
user = require_user(request)
row = db.conn().execute("SELECT muted FROM users WHERE id = ?", (user.user_id,)).fetchone()
if row and row["muted"]:
raise HTTPException(status_code=403, detail="Your account is muted")
if user.permission_state != "granted":
# 'pending' is the post-OTC waiting state; 'revoked' is the
# admin-undid-the-grant state. Both refuse with the same 403
# shape; the client distinguishes via `/api/auth/me` which
# carries `permission_state` in the response.
raise HTTPException(
status_code=403,
detail="Your beta access request is in review",
)
return user
+351
View File
@@ -0,0 +1,351 @@
"""§6.2 / v0.11.0: trust device for 30 days (roadmap item #9).
After a successful OTC or passcode sign-in, a contributor may check
"trust this device for 30 days." The framework then issues a
server-issued opaque token, hashes it (bcrypt) for storage in the
`device_trust` table, and sets a long-lived cookie carrying the raw
token. On a subsequent visit, the cookie is presented at
`/auth/device-trust/start`; if a non-expired, non-revoked row matches,
the session is re-established without another OTC / passcode round
trip.
The shape:
* `issue(user_id, user_agent)` mint a fresh CSPRNG token, hash it,
insert a row, and return the raw token + row id so the endpoint
can set the cookie. The 30-day expiry is the only knob; the
`revoked_at` column stays NULL.
* `lookup(raw_token)` walk the user's active rows (the unique
index keys on the hash, so we read a small candidate set), check
the bcrypt hash in constant time, drop any row whose `expires_at`
has passed or whose `revoked_at` is non-NULL, and return the
matched row or None. On a hit, refresh `last_seen_at`.
* `list_for_user(user_id)` return the active rows for the
/settings/devices surface. Revoked + expired rows are filtered out
so the surface only shows live trust grants.
* `revoke(user_id, row_id)` stamp `revoked_at` on the row. The
next lookup refuses the cookie token (the row is dead).
* `revoke_all(user_id)` bulk-revoke every active row for the user.
The /settings/devices surface's "revoke all" button calls this.
Cookie shape: `rfc_device_trust`. HttpOnly, Secure, SameSite=Lax,
Max-Age=2592000 (30 days), Path=/. The cookie value is the raw token;
server-side storage is the hash. The cookie is "essential" per the
v0.13.0 cookie-consent banner (it is part of authentication, not
analytics), so the framework sets it regardless of analytics /
other-cookies choices.
Constant-time comparison: bcrypt's `checkpw` is already constant-time
over the hash bytes. We walk the candidate set linearly with `_check`
which delegates to `bcrypt.checkpw`; no early-exit shortcut leaks
which row was the match.
The raw token never appears in a log line or an exception message;
the helpers carry the token only as a parameter and forget it after
hashing.
The cookie sits orthogonal to the §6.1 `permission_state` gate: a
revoked or pending user with a valid device-trust cookie still
re-establishes their session (the cookie identifies the user, not
their admission state), and the existing `require_contributor` /
`require_admin` dependencies in `auth.py` continue to refuse the
unrelated write surfaces.
"""
from __future__ import annotations
import logging
import secrets
from dataclasses import dataclass
import bcrypt
from . import db
from .auth import SessionUser
log = logging.getLogger(__name__)
# ---------------------------------------------------------------------------
# Tunables — hard-coded in v0.11.0 (§19.2 candidate to env-ify later).
# ---------------------------------------------------------------------------
TRUST_DURATION_DAYS = 30
COOKIE_NAME = "rfc_device_trust"
COOKIE_MAX_AGE_SECONDS = TRUST_DURATION_DAYS * 24 * 60 * 60
# 256 bits of CSPRNG entropy. `secrets.token_urlsafe(32)` yields ~43
# URL-safe characters; the bcrypt hash is what's stored, so the raw
# token only ever lives in the cookie.
TOKEN_BYTES = 32
# User-Agent header values seen in the wild can be unbounded; clamp
# to a reasonable ceiling so a hostile UA doesn't bloat the row.
USER_AGENT_MAX_LENGTH = 1024
# ---------------------------------------------------------------------------
# Issue
# ---------------------------------------------------------------------------
@dataclass
class IssueOutcome:
"""The shape returned from `issue`.
`raw_token` is the cookie value to send to the client; it never
appears in storage. `row_id` is the surrogate key for the
/settings/devices UI to address the row by id.
"""
raw_token: str
row_id: int
def _new_token() -> str:
return secrets.token_urlsafe(TOKEN_BYTES)
def _hash(token: str) -> str:
return bcrypt.hashpw(token.encode("utf-8"), bcrypt.gensalt()).decode("ascii")
def _check(token: str, token_hash: str) -> bool:
try:
return bcrypt.checkpw(token.encode("utf-8"), token_hash.encode("ascii"))
except (ValueError, TypeError):
return False
def _trim_user_agent(ua: str) -> str:
ua = (ua or "").strip()
if len(ua) > USER_AGENT_MAX_LENGTH:
return ua[:USER_AGENT_MAX_LENGTH]
return ua
def issue(user_id: int, user_agent: str) -> IssueOutcome:
"""Mint a fresh device-trust token + row for `user_id`.
The row's expiry is set 30 days in the future. The hash, not the
raw token, lands in the database. The caller (the endpoint) sets
the cookie with the raw token returned here.
"""
raw = _new_token()
h = _hash(raw)
ua = _trim_user_agent(user_agent)
cur = db.conn().execute(
f"""
INSERT INTO device_trust (user_id, device_token_hash, expires_at, user_agent)
VALUES (?, ?, datetime('now', '+{TRUST_DURATION_DAYS} days'), ?)
""",
(user_id, h, ua),
)
row_id = cur.lastrowid
return IssueOutcome(raw_token=raw, row_id=row_id)
# ---------------------------------------------------------------------------
# Lookup
# ---------------------------------------------------------------------------
@dataclass
class LookupOutcome:
"""The result of `lookup`.
`user` is populated only on a hit. `reason` distinguishes the
failure modes so the endpoint can decide whether to clear the
cookie ('expired', 'revoked', 'unknown') or just refuse ('invalid').
"""
ok: bool
user: SessionUser | None
reason: str # 'ok' | 'invalid' | 'unknown' | 'expired' | 'revoked'
row_id: int | None = None
def lookup(raw_token: str) -> LookupOutcome:
"""Resolve a presented cookie token to a user.
A hit refreshes `last_seen_at` on the matched row. A miss returns
a reason so the endpoint can clear the stale cookie if the row
was revoked or expired (vs. simply unknown, which probably means
the cookie was forged or the row was wiped by a /settings/devices
revoke from another browser).
"""
raw = (raw_token or "").strip()
if not raw:
return LookupOutcome(ok=False, user=None, reason="invalid")
# The unique index on `device_token_hash` would let us SELECT by
# hash if bcrypt were a stable hash, but bcrypt incorporates a
# per-row salt — equal tokens produce different hashes. We walk
# the candidate set instead. In practice the set is small (a
# human has a handful of trusted devices) and bcrypt is cheap on
# the order of milliseconds; the walk is bounded by the user's
# active device count.
#
# We don't pre-filter by `revoked_at IS NULL` here so that a
# token presented for a recently-revoked row produces a
# 'revoked' outcome (the endpoint surfaces a different shape).
# Same for expired: we let the walk hit and classify after.
rows = db.conn().execute(
"""
SELECT id, user_id, device_token_hash, expires_at, revoked_at
FROM device_trust
ORDER BY id DESC
""",
).fetchall()
matched = None
for row in rows:
if _check(raw, row["device_token_hash"]):
matched = row
break
if matched is None:
return LookupOutcome(ok=False, user=None, reason="unknown")
if matched["revoked_at"] is not None:
return LookupOutcome(ok=False, user=None, reason="revoked", row_id=matched["id"])
expired = db.conn().execute(
"SELECT datetime(?) < datetime('now') AS expired",
(matched["expires_at"],),
).fetchone()["expired"]
if expired:
return LookupOutcome(ok=False, user=None, reason="expired", row_id=matched["id"])
# Refresh last-seen so the /settings/devices surface can show the
# user when each device was last active. This is the only write
# the lookup path does on the hot read.
db.conn().execute(
"UPDATE device_trust SET last_seen_at = datetime('now') WHERE id = ?",
(matched["id"],),
)
user_row = db.conn().execute(
"""
SELECT id, gitea_id, gitea_login, email, display_name, avatar_url, role, permission_state
FROM users
WHERE id = ?
""",
(matched["user_id"],),
).fetchone()
if user_row is None:
# The user row was deleted but the device_trust row hadn't
# cascaded yet (shouldn't happen under the FK ON DELETE
# CASCADE — be defensive anyway). Treat as 'unknown' so the
# endpoint clears the cookie.
return LookupOutcome(ok=False, user=None, reason="unknown", row_id=matched["id"])
# Also stamp last_seen_at on the user row so the user's overall
# activity stamp keeps pace with cookie-only sign-ins.
db.conn().execute(
"UPDATE users SET last_seen_at = datetime('now') WHERE id = ?",
(matched["user_id"],),
)
return LookupOutcome(
ok=True,
user=SessionUser(
user_id=user_row["id"],
gitea_id=user_row["gitea_id"] or 0,
gitea_login=user_row["gitea_login"] or "",
display_name=user_row["display_name"],
email=user_row["email"] or "",
avatar_url=user_row["avatar_url"] or "",
role=user_row["role"],
permission_state=user_row["permission_state"] or "granted",
),
reason="ok",
row_id=matched["id"],
)
# ---------------------------------------------------------------------------
# List / revoke (for the /settings/devices surface)
# ---------------------------------------------------------------------------
@dataclass
class DeviceRow:
"""The shape the /settings/devices endpoint returns.
Note the absence of `device_token_hash` the hash is structurally
private, and the surface has no use for it.
"""
id: int
created_at: str
expires_at: str
last_seen_at: str
user_agent: str
def list_for_user(user_id: int) -> list[DeviceRow]:
"""Active device-trust rows for the user, freshest first.
Filters out revoked rows and rows whose expiry has passed; the
surface only shows live trust grants. A user wondering "which
devices are signed in" gets the answer that matches what the
framework would actually accept on a presented cookie.
"""
rows = db.conn().execute(
"""
SELECT id, created_at, expires_at, last_seen_at, user_agent
FROM device_trust
WHERE user_id = ?
AND revoked_at IS NULL
AND datetime(expires_at) > datetime('now')
ORDER BY last_seen_at DESC, id DESC
""",
(user_id,),
).fetchall()
return [
DeviceRow(
id=row["id"],
created_at=row["created_at"],
expires_at=row["expires_at"],
last_seen_at=row["last_seen_at"],
user_agent=row["user_agent"] or "",
)
for row in rows
]
def revoke(user_id: int, row_id: int) -> bool:
"""Revoke a single device-trust row for the given user.
Returns True iff a row was matched (still active, belongs to the
user). The user-id scope is enforced in SQL so a hostile client
cannot revoke another user's row by guessing ids.
"""
cur = db.conn().execute(
"""
UPDATE device_trust
SET revoked_at = datetime('now')
WHERE id = ?
AND user_id = ?
AND revoked_at IS NULL
""",
(row_id, user_id),
)
return cur.rowcount > 0
def revoke_all(user_id: int) -> int:
"""Revoke every active device-trust row for the user. Returns the
count of rows touched.
The /settings/devices "revoke all" button calls this. The user's
current request stays authenticated via its session cookie; the
device-trust cookie on the current device is also revoked, but
the session middleware's `rfc_session` cookie keeps the request
flow alive until the user signs out or the session cookie
expires.
"""
cur = db.conn().execute(
"""
UPDATE device_trust
SET revoked_at = datetime('now')
WHERE user_id = ?
AND revoked_at IS NULL
""",
(user_id,),
)
return cur.rowcount
+61
View File
@@ -0,0 +1,61 @@
"""User-facing docs source.
Mirrors `philosophy.py` shape. Serves `DOCS.md` from the repo root
the framework's plain-prose user guide to roles, contribution flow,
and notification surfaces, distinct from the binding `SPEC.md`. Read
from disk on first call and cached in-process; the periodic
reconciler can call `refresh()` to pick up out-of-band edits.
`DOCS_PATH` overrides the default location if a deployment hosts the
file elsewhere (a meta-repo working-tree clone, a sync target, etc.).
"""
from __future__ import annotations
import logging
import os
import threading
from pathlib import Path
log = logging.getLogger(__name__)
_DEFAULT_PATH = Path(__file__).resolve().parents[2] / "DOCS.md"
_lock = threading.Lock()
_cache: dict | None = None
def _resolved_path() -> Path:
override = os.environ.get("DOCS_PATH", "").strip()
if override:
return Path(override).expanduser().resolve()
return _DEFAULT_PATH
def load(force: bool = False) -> dict:
"""Return the cached `{body, path, mtime}` payload, reading from disk
on first call or when `force=True`.
"""
global _cache
with _lock:
if _cache is not None and not force:
return _cache
path = _resolved_path()
try:
text = path.read_text(encoding="utf-8")
mtime = path.stat().st_mtime
except FileNotFoundError:
log.warning("DOCS.md not found at %s — serving placeholder", path)
text = (
"# DOCS.md not found\n\n"
"The deployment is missing its user guide. Set "
"DOCS_PATH or place DOCS.md at the project root."
)
mtime = 0.0
_cache = {"body": text, "path": str(path), "mtime": mtime}
return _cache
def refresh() -> dict:
"""Force-reread from disk. Returns the new payload."""
return load(force=True)
+11
View File
@@ -139,6 +139,10 @@ _EVENT_TO_CATEGORY: dict[str, str] = {
"graduation_complete": "personal-direct",
"super_draft_graduation_ready": "admin-actionable",
"claim_opened": "structural",
# v0.9.0: roadmap item #7. A fresh beta-access request lands as
# an admin-actionable signal so it consults `email_admin_actionable`
# and reaches owners/admins only.
"new_beta_request": "admin-actionable",
}
@@ -285,6 +289,13 @@ def _deep_link(payload: dict, cfg: EmailConfig) -> str:
slug = payload.get("rfc_slug")
pr = payload.get("pr_number")
branch = payload.get("branch_name")
event_kind = payload.get("event_kind")
# v0.9.0: framework-scoped admin signals link to the admin
# surface, not /rfc/... The `new_beta_request` event is the
# canonical example; future framework-scoped admin events
# may reuse the same branch.
if event_kind == "new_beta_request":
return f"{cfg.app_url}/admin/users"
if slug and pr:
return f"{cfg.app_url}/rfc/{slug}/pr/{pr}"
if slug and branch:
+186 -6
View File
@@ -10,8 +10,8 @@ import logging
import secrets
from contextlib import asynccontextmanager
from fastapi import APIRouter, FastAPI, HTTPException, Request
from fastapi.responses import RedirectResponse
from fastapi import APIRouter, FastAPI, HTTPException, Request, Response
from fastapi.responses import JSONResponse, RedirectResponse
from pydantic import BaseModel, Field
from starlette.middleware.sessions import SessionMiddleware
@@ -20,12 +20,14 @@ from . import (
auth,
cache,
db,
device_trust as device_trust_mod,
digest,
email_otc,
hygiene,
otc,
passcode as passcode_mod,
providers as providers_mod,
turnstile,
webhooks,
)
from .bot import Bot
@@ -38,11 +40,25 @@ log = logging.getLogger("rfc_app")
class OtcRequestBody(BaseModel):
email: str = Field(min_length=3, max_length=320)
# v0.12.0 / roadmap item #10: CloudFlare Turnstile token from the
# frontend widget. Optional in the body so a deployment that has
# not yet wired the Turnstile site key (or a dev environment with
# the widget intentionally skipped) still routes through the same
# endpoint; the backend turnstile.verify_token call decides whether
# to admit the request based on `TURNSTILE_REQUIRED` + presence of
# the secret.
turnstile_token: str | None = Field(default=None, max_length=4096)
class OtcVerifyBody(BaseModel):
email: str = Field(min_length=3, max_length=320)
code: str = Field(min_length=1, max_length=16)
# v0.11.0 — "trust this device for 30 days" checkbox on the Login.jsx
# OTC step. When true and verify succeeds, the server issues a fresh
# device-trust row and sets the `rfc_device_trust` cookie on the
# response. Defaults to false so existing clients that don't send
# the flag continue to behave the way they did pre-v0.11.0.
trust_device: bool = False
class PasscodeSetBody(BaseModel):
@@ -52,6 +68,8 @@ class PasscodeSetBody(BaseModel):
class PasscodeVerifyBody(BaseModel):
email: str = Field(min_length=3, max_length=320)
passcode: str = Field(min_length=1, max_length=64)
# v0.11.0 — same trust-device opt-in as the OTC verify body.
trust_device: bool = False
@asynccontextmanager
@@ -117,6 +135,48 @@ def create_app() -> FastAPI:
app = create_app()
def _set_device_trust_cookie(response: Response, raw_token: str) -> None:
"""Attach the v0.11.0 device-trust cookie to the response.
HttpOnly + Secure + SameSite=Lax + 30-day Max-Age + Path=/. The
cookie value is the raw token; server-side storage is the hash.
The cookie is "essential" per the v0.13.0 cookie-consent contract
(it is part of authentication), so we set it regardless of the
user's analytics / other-cookies choice.
Secure=True means the cookie is only ever sent over HTTPS. The
SessionMiddleware in `create_app` keeps `https_only=False` for
dev parity, but the device-trust cookie holds a 30-day credential
and must not travel cleartext production deployments serve over
HTTPS, so Secure on the device-trust cookie is non-negotiable.
"""
response.set_cookie(
key=device_trust_mod.COOKIE_NAME,
value=raw_token,
max_age=device_trust_mod.COOKIE_MAX_AGE_SECONDS,
path="/",
secure=True,
httponly=True,
samesite="lax",
)
def _clear_device_trust_cookie(response: Response) -> None:
"""Delete the device-trust cookie on the response.
Used when the framework detects a presented cookie that is
expired, revoked, or otherwise stale the next request from
this device will not carry a dead token.
"""
response.delete_cookie(
key=device_trust_mod.COOKIE_NAME,
path="/",
secure=True,
httponly=True,
samesite="lax",
)
def _oauth_router(config) -> APIRouter:
router = APIRouter()
@@ -164,7 +224,27 @@ def _oauth_router(config) -> APIRouter:
# ---------------------------------------------------------------
@router.post("/auth/otc/request")
async def otc_request(body: OtcRequestBody):
async def otc_request(body: OtcRequestBody, request: Request):
# v0.12.0 / roadmap item #10: gate the request on a successful
# Turnstile siteverify before the bcrypt hash + SMTP send. The
# check runs first so a failed challenge spends no rate budget
# and produces no envelope. When the operator has not wired the
# secret AND TURNSTILE_REQUIRED=false (the default), the gate
# opens — see `backend/app/turnstile.py` for the full matrix.
client_ip = request.client.host if request.client else None
ts = turnstile.verify_token(body.turnstile_token, client_ip=client_ip)
if not ts.ok:
if ts.reason == "misconfigured":
# TURNSTILE_REQUIRED=true but the secret is unset. This
# is an operator/config problem, not a client problem;
# surface as 500 so the operator notices in their logs
# rather than blaming the user's browser.
raise HTTPException(500, "auth misconfigured")
# missing-token / failed / network → uniform 400 so the
# response does not enumerate which leg of the challenge
# broke. The reason is in the server logs.
raise HTTPException(400, "verification failed")
outcome = otc.request_code(body.email)
if outcome.reason == "cooldown":
# Loud failure per the rate-limit primitive — the abuse
@@ -177,11 +257,43 @@ def _oauth_router(config) -> APIRouter:
return {"ok": True}
@router.post("/auth/otc/verify")
async def otc_verify(body: OtcVerifyBody, request: Request):
async def otc_verify(body: OtcVerifyBody, request: Request, response: Response):
result = otc.verify_code(body.email, body.code)
if not result.ok or result.user is None:
raise HTTPException(400, "Invalid or expired code")
auth.store_session(request, result.user)
# v0.8.0: surface `needs_profile` so the Login.jsx surface can
# decide whether to advance to the first/last/why capture step
# or jump straight to "/". `needs_profile=true` iff the user
# is `permission_state='pending'` AND the row has no profile
# fields yet — a fresh OTC user. Grandfathered users
# (`permission_state='granted'`) and pending users who already
# captured their fields both read as false.
row = db.conn().execute(
"SELECT first_name, last_name, beta_request_reason FROM users WHERE id = ?",
(result.user.user_id,),
).fetchone()
first_name = (row["first_name"] if row else None) or ""
last_name = (row["last_name"] if row else None) or ""
beta_request_reason = (row["beta_request_reason"] if row else None) or ""
needs_profile = (
result.user.permission_state == "pending"
and not first_name
and not last_name
and not beta_request_reason
)
# v0.11.0 — opt-in device trust. The checkbox lives on the
# Login.jsx OTC step; when true, the server mints a fresh
# device-trust row and sets the long-lived cookie. The cookie
# is "essential" per the v0.13.0 consent contract (it is part
# of authentication, not analytics) so it lands regardless of
# the user's analytics / other-cookies choice. We capture the
# User-Agent at issuance so the /settings/devices surface can
# render a rough device label.
if body.trust_device:
ua = request.headers.get("user-agent", "")
outcome = device_trust_mod.issue(result.user.user_id, ua)
_set_device_trust_cookie(response, outcome.raw_token)
return {
"ok": True,
"user": {
@@ -189,7 +301,9 @@ def _oauth_router(config) -> APIRouter:
"display_name": result.user.display_name,
"email": result.user.email,
"role": result.user.role,
"permission_state": result.user.permission_state,
},
"needs_profile": needs_profile,
}
# ---------------------------------------------------------------
@@ -232,12 +346,16 @@ def _oauth_router(config) -> APIRouter:
return {"ok": True}
@router.post("/auth/passcode/verify")
async def passcode_verify(body: PasscodeVerifyBody, request: Request):
async def passcode_verify(body: PasscodeVerifyBody, request: Request, response: Response):
"""Sign in with email + passcode. Returns the standard session
payload on success; HTTP 423 with `locked_until` when the
account is in the lockout window; HTTP 400 for every other
failure (the wrong-vs-unknown distinction is intentionally
collapsed so a probing client cannot enumerate emails)."""
collapsed so a probing client cannot enumerate emails).
v0.11.0: the body's `trust_device` flag, if true, mints a
fresh device-trust row and sets the long-lived cookie. Same
opt-in contract as `/auth/otc/verify`."""
result = passcode_mod.verify_passcode(body.email, body.passcode)
if result.reason == "locked":
raise HTTPException(
@@ -250,6 +368,10 @@ def _oauth_router(config) -> APIRouter:
if not result.ok or result.user is None:
raise HTTPException(400, "Invalid passcode")
auth.store_session(request, result.user)
if body.trust_device:
ua = request.headers.get("user-agent", "")
outcome = device_trust_mod.issue(result.user.user_id, ua)
_set_device_trust_cookie(response, outcome.raw_token)
return {
"ok": True,
"user": {
@@ -260,4 +382,62 @@ def _oauth_router(config) -> APIRouter:
},
}
# ---------------------------------------------------------------
# v0.11.0: trust device for 30 days (§6.2, roadmap item #9).
#
# The /auth/device-trust/start endpoint resolves a presented
# `rfc_device_trust` cookie. If it matches a non-expired,
# non-revoked row, the session is re-established and the client
# is told to skip OTC/passcode entry. A stale cookie (expired or
# revoked) is cleared on the response. A miss is structurally
# silent — the client falls back to the email step.
#
# The endpoint is anonymous-reachable: a returning visitor with
# the cookie hits this before the email step. We do not gate it
# on a session because the entire point is to establish one.
# ---------------------------------------------------------------
@router.post("/auth/device-trust/start")
async def device_trust_start(request: Request):
"""Sign in via a presented device-trust cookie.
On a hit, re-establishes the session in the cookie store and
returns a user payload shaped like /auth/otc/verify (minus
`needs_profile`, which a returning device-trust user is
structurally past they signed in at least once before).
On a miss, returns 401 + clears the stale cookie. An
'unknown' miss (cookie present but no row matches) also
clears, since the token is dead to the server either way.
Note on response construction: we return a `JSONResponse`
directly rather than raising `HTTPException` on the miss
path because FastAPI's exception handler builds a new
response from scratch and would drop any `set_cookie` /
`delete_cookie` calls. The hand-built `JSONResponse` lets
us attach the cookie-clear header alongside the 401.
"""
raw = request.cookies.get(device_trust_mod.COOKIE_NAME, "")
if not raw:
return JSONResponse({"detail": "No device trust"}, status_code=401)
outcome = device_trust_mod.lookup(raw)
if not outcome.ok or outcome.user is None:
# Clear the stale cookie so subsequent requests don't
# keep replaying a dead token. We surface 401 in all
# cases so a probing client can't tell "your row was
# revoked" from "this token never existed".
response = JSONResponse({"detail": "Device trust invalid"}, status_code=401)
_clear_device_trust_cookie(response)
return response
auth.store_session(request, outcome.user)
return {
"ok": True,
"user": {
"id": outcome.user.user_id,
"display_name": outcome.user.display_name,
"email": outcome.user.email,
"role": outcome.user.role,
"permission_state": outcome.user.permission_state,
},
}
return router
+72
View File
@@ -64,6 +64,7 @@ log = logging.getLogger(__name__)
CATEGORY_PERSONAL = "personal-direct"
CATEGORY_STRUCTURAL = "structural"
CATEGORY_CHURN = "churn"
CATEGORY_ADMIN_ACTIONABLE = "admin-actionable"
# Action kinds whose actor's first interaction with a slug triggers
# auto-watch per §15.6. The substantive-gesture list in the spec is
@@ -208,6 +209,67 @@ def fan_out_from_action(
)
def fan_out_new_beta_request(
*,
requester_user_id: int,
) -> None:
"""v0.9.0 (roadmap item #7): announce a fresh beta-access request to
every admin/owner.
Called from `POST /api/auth/me/beta-request` after the row's
first/last/why fields are populated. Fan-out shape mirrors the §15
chokepoint contract: one row per recipient, written via `_emit_one`
so the SSE broadcast + email dispatch run through the same surface
every other notification uses. The event has no rfc_slug (it is
framework-scoped, not RFC-scoped); the deep-link payload points
`/admin/users` instead of `/rfc/<slug>`.
Actor is the requester per §15.9 (the underlying user, never the
bot). Category is `admin-actionable` so the §15.4 email gate
consults `email_admin_actionable` (owners/admins-only by
construction) and the digest exclusion rules treat it identically
to other admin-actionable signals (graduation_ready et al).
Recipients are owners + admins minus the requester themselves
(a self-promotion shouldn't reach the requester's own inbox). The
requester is never in the role set in practice the endpoint
refuses 'granted'/'revoked' callers and a fresh OTC user lands
`contributor`+`pending` but we filter regardless so the call
is robust to future changes in the auth gate.
"""
requester = db.conn().execute(
"SELECT first_name, last_name, email, display_name FROM users WHERE id = ?",
(requester_user_id,),
).fetchone()
if requester is None:
return
first = (requester["first_name"] or "").strip()
last = (requester["last_name"] or "").strip()
email = requester["email"] or ""
display = requester["display_name"] or email or "a new user"
full_name = (f"{first} {last}").strip() or display
details = {
"requester_user_id": requester_user_id,
"requester_first_name": first,
"requester_last_name": last,
"requester_email": email,
"requester_display": full_name,
}
for recipient_id in _admin_user_ids():
if recipient_id == requester_user_id:
continue
_emit_one(
recipient_user_id=recipient_id,
event_kind="new_beta_request",
category=CATEGORY_ADMIN_ACTIONABLE,
actor_user_id=requester_user_id,
rfc_slug=None,
branch_name=None,
pr_number=None,
details=details,
)
def fan_out_chat_message(
*,
actor_user_id: int,
@@ -707,6 +769,16 @@ def render_summary(event_kind: str, actor_display: str | None, rfc_title: str |
return f"{actor} began graduating {title}."
if event_kind == "pr_conflict_with_main":
return f"{actor} started a resolution branch on {title}."
if event_kind == "new_beta_request":
# v0.9.0: framework-scoped, not RFC-scoped. The actor (the
# requester) and the captured full name + email read as
# one self-contained sentence; the inbox row and the email
# body share this text per §15.4.
full_name = extras.get("requester_display") or actor
email_addr = extras.get("requester_email") or ""
if email_addr:
return f"New beta-access request from {full_name} ({email_addr})."
return f"New beta-access request from {full_name}."
return f"{event_kind} on {title}"
+52 -45
View File
@@ -1,4 +1,4 @@
"""§6.2 / v0.7.0: email + one-time-code sign-in.
"""§6.2 / v0.7.0 / v0.8.0: email + one-time-code sign-in.
Replaces the Gitea OAuth gesture as the primary human-auth path. The
Gitea bot user + token are still needed for server-side git
@@ -25,14 +25,25 @@ The shape:
`users` row already carries `email` (case-insensitive), it is
reused `gitea_id` is left alone so a grandfathered OAuth-era
user keeps the linker intact. Otherwise a fresh contributor
row is provisioned with `gitea_id = NULL`, `gitea_login = NULL`.
row is provisioned with `gitea_id = NULL`, `gitea_login = NULL`,
and `permission_state = 'pending'` (v0.8.0 see below).
The endpoints in `main.py` thin-wrap this module. The allowlist gate
from v0.3.0 is consulted at request time if `allowed_emails` is
populated and the requested address isn't on it, the request returns
202 as usual but no email is sent. This intentionally does not leak
allowlist state to the caller; the §19.2 candidate for v0.8.0
replaces this gate with an admin-grant flow.
The endpoints in `main.py` thin-wrap this module.
v0.8.0 (roadmap item #6) replaces the v0.3.0 `allowed_emails` gate at
the request surface. The request handler used to silently drop OTC
requests for emails not on the allowlist; now any valid email
receives a code. The admission gate moves to `permission_state` on
the freshly-provisioned `users` row: a fresh user lands in 'pending'
and waits for an admin grant before write endpoints accept them.
Read surfaces stay open (the same blast radius v0.6.0 / item #4
already audited for anonymous viewers).
The `allowed_emails` table itself stays in the schema as a
fast-path bypass the admin UI from v0.3.0 continues to manage it,
and a future release (v0.9.0's admin user-management page) collapses
the two admission surfaces into one. The OTC request path no
longer consults the table.
"""
from __future__ import annotations
@@ -44,7 +55,7 @@ from dataclasses import dataclass
import bcrypt
from . import db
from .auth import SessionUser, allowlist_is_active
from .auth import SessionUser
log = logging.getLogger(__name__)
@@ -101,25 +112,15 @@ def _check_code(code: str, code_hash: str) -> bool:
return False
# ---------------------------------------------------------------------------
# Allowlist gate — shared with the OAuth flow.
# ---------------------------------------------------------------------------
def _allowlist_admits(email: str) -> bool:
"""The same allowlist v0.3.0 introduced for OAuth, applied to OTC
requests. If the allowlist is populated and the email is not on it,
we still respond 202 to the caller, but no code is sent."""
if not allowlist_is_active():
return True
row = db.conn().execute(
"SELECT 1 FROM allowed_emails WHERE email = ? LIMIT 1", (email,)
).fetchone()
return row is not None
# ---------------------------------------------------------------------------
# Request path
#
# v0.8.0: the allowlist gate from v0.7.0 / v0.3.0 is removed here. Any
# valid email receives a code; the admission gate moved to
# `permission_state` on the freshly-provisioned `users` row (see
# `provision_or_link_user`). The `allowed_emails` table stays in the
# schema (admin UI from v0.3.0 still manages it); v0.9.0's admin
# user-management page will collapse the two surfaces.
# ---------------------------------------------------------------------------
@@ -127,14 +128,16 @@ def _allowlist_admits(email: str) -> bool:
class RequestOutcome:
"""The outcome of a `request_code` call.
`code` is None whenever no code was generated either because the
allowlist denied the email or because the cooldown window blocked
the request. The caller (the API endpoint) does not surface this
distinction to the user; it returns 202 either way.
`code` is None whenever no code was generated the cooldown
window blocked the request or the email was syntactically
invalid. The caller (the API endpoint) does not surface the
invalid-email shape to the user; it returns 202 either way.
The cooldown shape surfaces as a loud 429 per the v0.7.0
contract.
"""
sent: bool
code: str | None
reason: str # 'sent' | 'allowlist' | 'cooldown' | 'invalid'
reason: str # 'sent' | 'cooldown' | 'invalid'
def request_code(email: str) -> RequestOutcome:
@@ -160,13 +163,6 @@ def request_code(email: str) -> RequestOutcome:
if row is not None:
return RequestOutcome(sent=False, code=None, reason="cooldown")
# Allowlist: silently drop the send if the email isn't on the list.
# The row is not written either — there's nothing for verify to
# match against, so the user-facing experience is "I never got an
# email", which is the intended shape for the private-beta gate.
if not _allowlist_admits(email):
return RequestOutcome(sent=False, code=None, reason="allowlist")
# Invalidate prior unused codes for this email. A re-request is
# always for the most recent code; older codes are dead.
db.conn().execute(
@@ -275,12 +271,16 @@ def provision_or_link_user(email: str) -> SessionUser:
1. An existing row whose email equals (case-insensitive) the
requested email the OAuth-era user is grandfathered in via
this path. `gitea_id` is preserved so a future OAuth round
trip still resolves the same row.
trip still resolves the same row. `permission_state` is
read off the row as-is grandfathered users come through
migration with 'granted' (the column default), so their
contributor capabilities are unaffected.
2. Otherwise: a fresh contributor row with `gitea_id = NULL`,
`gitea_login = NULL`. The display name defaults to the local
part of the email (everything before the `@`) users can
rename later via the §19.2 first-OTC profile-capture flow
that v0.8.0 introduces.
`gitea_login = NULL`, and `permission_state = 'pending'`
(v0.8.0). The display name defaults to the local part of
the email (everything before the `@`); a separate
`POST /auth/me/beta-request` call lands first name / last
name / "why I want access" on the same row.
The §6.1 owner-zero bootstrap still applies: if the email matches
the configured `OWNER_GITEA_LOGIN`-derived owner identity, the row
@@ -307,13 +307,19 @@ def provision_or_link_user(email: str) -> SessionUser:
email=existing["email"] or email,
avatar_url=existing["avatar_url"] or "",
role=existing["role"],
permission_state=existing["permission_state"] or "granted",
)
display = email.split("@", 1)[0] or email
# v0.8.0: 'pending' is the explicit insert value; the migration
# default of 'granted' is what passes grandfathered users
# through. A fresh OTC user lands in 'pending' regardless of
# what the migration default says, so the gate engages reliably
# even if a future migration changes the default.
cur = db.conn().execute(
"""
INSERT INTO users (gitea_id, gitea_login, email, display_name, avatar_url, role)
VALUES (NULL, NULL, ?, ?, '', 'contributor')
INSERT INTO users (gitea_id, gitea_login, email, display_name, avatar_url, role, permission_state)
VALUES (NULL, NULL, ?, ?, '', 'contributor', 'pending')
""",
(email, display),
)
@@ -326,4 +332,5 @@ def provision_or_link_user(email: str) -> SessionUser:
email=email,
avatar_url="",
role="contributor",
permission_state="pending",
)
+148
View File
@@ -0,0 +1,148 @@
"""§6.2 / v0.12.0 / roadmap item #10: CloudFlare Turnstile siteverify.
The OTC request endpoint (`/auth/otc/request`) is the abuse hot path
of the auth surface since v0.7.0 the per-email cooldown stops the
trivial back-to-back loop, but it does not stop a distributed scraper
that fans out across a large invitee list to harvest the "this email
is admitted vs. this email is not" signal indirectly (timing
differences, SMTP bounce-rate observation). v0.12.0 gates the request
endpoint behind a one-step browser-side Turnstile challenge before the
bcrypt hash + SMTP send.
Stateless: no DB writes, no schema change. The siteverify call to
CloudFlare lives entirely in this module; the endpoint handler in
`main.py` thin-wraps `verify_token`.
Tunables (read at call time so tests can monkeypatch):
* `CLOUDFLARE_TURNSTILE_SECRET` the operator-provisioned secret
key from the Turnstile dashboard. Lives in GCP Secret Manager in
production; absent in tests (which monkeypatch the siteverify
transport). When unset, the behavior depends on `TURNSTILE_REQUIRED`:
- `TURNSTILE_REQUIRED=true` fail closed (`misconfigured`).
- `TURNSTILE_REQUIRED=false` (default) skip verification entirely
and admit the request. This is the dev/test path and the
"operator hasn't wired the secret yet" path; production
deployments **should** set `TURNSTILE_REQUIRED=true` once the
secret is in place so a regression in the secret wiring fails
loudly instead of silently disabling abuse defense.
* `TURNSTILE_REQUIRED` `true` / `false` (default `false`).
When `false` and the secret is absent, the gate is open. When
`true` and the secret is absent, the endpoint refuses with a
misconfigured-auth shape rather than silently letting requests
through.
* `TURNSTILE_SITEVERIFY_URL` points at the real CloudFlare
endpoint by default. Override in tests to redirect at a mock
URL when `httpx.MockTransport` isn't ergonomic for the case.
The siteverify contract is documented at
https://developers.cloudflare.com/turnstile/get-started/server-side-validation/.
We POST `secret` + `response` (and optionally `remoteip`) as form
fields and read back `{"success": true|false, ...}`. Any network /
parse failure on the siteverify call is treated as a verification
failure (`network`) the abuse path is to skip the challenge, so
"can't reach CloudFlare" defaults to "refuse the request" when
`TURNSTILE_REQUIRED=true`, and "admit" when `TURNSTILE_REQUIRED=false`.
"""
from __future__ import annotations
import logging
import os
from dataclasses import dataclass
import httpx
log = logging.getLogger(__name__)
SITEVERIFY_URL = "https://challenges.cloudflare.com/turnstile/v0/siteverify"
def _secret() -> str:
return os.environ.get("CLOUDFLARE_TURNSTILE_SECRET", "").strip()
def _required() -> bool:
raw = os.environ.get("TURNSTILE_REQUIRED", "").strip().lower()
return raw in ("1", "true", "yes", "on")
def _siteverify_url() -> str:
return os.environ.get("TURNSTILE_SITEVERIFY_URL", "").strip() or SITEVERIFY_URL
@dataclass
class VerifyOutcome:
"""Result of a Turnstile siteverify call.
`ok`: the request **may proceed**. True both for "siteverify said
success" and for "no secret configured AND not required" (the
dev/test soft-fail path).
`reason`: one of
* 'ok' siteverify returned success.
* 'skipped' no secret configured, TURNSTILE_REQUIRED=false.
The gate is open; the endpoint admits the request.
* 'misconfigured' TURNSTILE_REQUIRED=true but no secret in env.
The endpoint fails closed with 500.
* 'missing-token' the client did not send a token at all and
verification is required.
* 'failed' siteverify returned success=false. The
endpoint refuses with 400.
* 'network' siteverify call raised. Treated as a failure
under TURNSTILE_REQUIRED=true.
"""
ok: bool
reason: str
def verify_token(token: str | None, *, client_ip: str | None = None) -> VerifyOutcome:
"""Validate a Turnstile token against CloudFlare's siteverify endpoint.
Returns a VerifyOutcome describing whether the calling endpoint
should proceed. The endpoint maps `ok=False` to an HTTP status per
the `reason`:
* 'misconfigured' 500 "auth misconfigured"
* 'missing-token' / 'failed' / 'network' 400 "verification failed"
Tests monkeypatch `httpx.post` (or set `TURNSTILE_SITEVERIFY_URL`
+ a MockTransport client) to avoid touching the real CloudFlare
endpoint. No real keys are ever embedded in tests.
"""
secret = _secret()
required = _required()
if not secret:
if required:
log.warning("Turnstile required but CLOUDFLARE_TURNSTILE_SECRET is unset; failing closed")
return VerifyOutcome(ok=False, reason="misconfigured")
# Dev/test/soft-fail path: no secret, not required → gate is open.
return VerifyOutcome(ok=True, reason="skipped")
if not token or not token.strip():
# Secret is set, so verification is in force. A missing token
# is a hard refuse — the frontend should have rendered the
# widget and collected one.
return VerifyOutcome(ok=False, reason="missing-token")
data = {"secret": secret, "response": token.strip()}
if client_ip:
data["remoteip"] = client_ip
try:
response = httpx.post(_siteverify_url(), data=data, timeout=10.0)
payload = response.json()
except Exception as exc: # network, JSON parse, etc.
log.warning("Turnstile siteverify call failed: %s", exc)
return VerifyOutcome(ok=False, reason="network")
if payload.get("success") is True:
return VerifyOutcome(ok=True, reason="ok")
# `error-codes` is a list of strings on failure; we log the codes
# for the operator without surfacing them to the client.
log.info("Turnstile siteverify rejected token: %s", payload.get("error-codes"))
return VerifyOutcome(ok=False, reason="failed")
+76
View File
@@ -0,0 +1,76 @@
-- §6.1 / §6.2 / §14.1 / v0.8.0: open beta-access request flow (roadmap item #6).
--
-- This release replaces v0.3.0's `allowed_emails` allowlist as the
-- admission control. Anyone with a valid email can sign in via the
-- v0.7.0 OTC flow; a fresh user lands in `permission_state='pending'`
-- until an admin grants access. The first-OTC flow captures three
-- profile fields (first name, last name, free-text "why I should be
-- included in the beta") that the admin sees when triaging the
-- request queue. The `allowed_emails` table stays in the schema as a
-- fast-path bypass — populated rows are still readable by the
-- existing admin UI; the OTC `/request` handler no longer consults
-- it. v0.9.0's admin user-management page will replace the
-- allowlist UI entirely.
--
-- Schema additions:
--
-- * `permission_state` — three-state CHECK: 'pending' | 'granted' |
-- 'revoked'. Default 'granted' so every row at migration time
-- passes through unaffected; only newly provisioned OTC users
-- land in 'pending' (the OTC verify path sets the column
-- explicitly on a fresh row, per `app/otc.py`). 'revoked' is the
-- admin gesture for an account that earned a grant then later
-- lost it; v0.8.0 doesn't surface a revoke UI, but the schema
-- slot is here so v0.9.0's admin user-management page can flip
-- the column without another migration.
--
-- * `first_name`, `last_name` — nullable TEXT. Captured on the
-- first OTC sign-in via `POST /auth/me/beta-request`. Existing
-- rows (OAuth-era users, OTC users provisioned in v0.7.0) carry
-- NULL through the migration; the admin queue treats an
-- unpopulated capture as "auto-grandfathered" since the row's
-- `permission_state` is already 'granted'.
--
-- * `beta_request_reason` — nullable TEXT. The free-text "why I
-- should be included" from the capture form. Bounded to ~4000
-- chars at the endpoint layer (no DB-level constraint —
-- SQLite's TEXT is unbounded).
--
-- * `permission_decided_by` — nullable INTEGER. The `users.id` of
-- the admin who flipped `permission_state` from 'pending' to
-- 'granted' (or 'granted' to 'revoked'). NULL for grandfathered
-- rows (they were never decided — they passed through at
-- migration). ON DELETE SET NULL because losing the admin row
-- should not cascade-delete the user whose access they granted.
--
-- * `permission_decided_at` — nullable TEXT timestamp (ISO 8601,
-- same shape as the existing `created_at` / `last_seen_at`).
-- Co-populated with `permission_decided_by` on each decision.
--
-- Grandfathered-row invariant:
--
-- Every row that exists at migration time has
-- `permission_state='granted'` and `permission_decided_by=NULL`
-- (the column default + NULL preservation). v0.8.0's auth gate
-- reads `permission_state='granted'` as the admission check, so
-- no existing user is locked out by the upgrade. v0.7.0's OTC
-- path is patched in the same release to set
-- `permission_state='pending'` explicitly on a fresh row, so the
-- gate engages only for users provisioned after the upgrade.
ALTER TABLE users ADD COLUMN permission_state TEXT NOT NULL DEFAULT 'granted'
CHECK (permission_state IN ('pending', 'granted', 'revoked'));
ALTER TABLE users ADD COLUMN first_name TEXT;
ALTER TABLE users ADD COLUMN last_name TEXT;
ALTER TABLE users ADD COLUMN beta_request_reason TEXT;
ALTER TABLE users ADD COLUMN permission_decided_by INTEGER
REFERENCES users(id) ON DELETE SET NULL;
ALTER TABLE users ADD COLUMN permission_decided_at TEXT;
-- Index for the v0.9.0 admin queue: list pending requests ordered by
-- when the user's row was created (the implicit "request received at"
-- timestamp, since v0.8.0 sets pending at the same moment as the row
-- itself is inserted via the OTC verify path).
CREATE INDEX idx_users_permission_state ON users (permission_state);
+75
View File
@@ -0,0 +1,75 @@
-- §6.2 / v0.11.0: trust device for 30 days (roadmap item #9).
--
-- After a successful OTC or passcode sign-in, the user can check
-- "trust this device for 30 days." The framework then issues a
-- server-issued opaque device-trust token, stores its hash on this
-- table, and sets a long-lived HttpOnly + Secure + SameSite=Lax
-- cookie carrying the raw token. On a subsequent visit, the cookie is
-- presented at `/auth/device-trust/start`; if the server can match the
-- hash to a non-expired non-revoked row, the user is signed in without
-- another OTC / passcode round-trip.
--
-- v0.11.0 introduces no new env vars. The 30-day window is hard-coded
-- in `backend/app/device_trust.py`; raising or lowering it (or making
-- it user-selectable) is a §19.2 candidate, alongside the cross-device
-- session-revocation surface this table will eventually share with the
-- v0.10.0 passcode-lockout shape (see SPEC §19.2 / SESSIONS-AND-DEVICES).
--
-- Storage shape:
--
-- * `id` — surrogate key. Lets the revoke-device UI address a single
-- row by id without leaking the token shape.
-- * `user_id` — FK into users(id) with cascade on delete. A deleted
-- user automatically loses every trusted device.
-- * `device_token_hash` — bcrypt hash of the random opaque token
-- issued at trust-time. The raw token only ever lives in the
-- outbound `Set-Cookie` header and the inbound `Cookie` header;
-- server-side storage is the hash, so a DB compromise does not
-- hand attackers a stash of valid device tokens.
-- * `created_at` — when the row was issued.
-- * `expires_at` — `created_at + 30 days`. A row past this timestamp
-- is dead; the lookup path refuses it without further checks.
-- * `user_agent` — the User-Agent header captured at issuance.
-- Stored verbatim (truncated to 1024 chars at the application
-- layer) so the revoke-device UI can show a rough device label.
-- Not used for any auth decision — purely a hint to the user
-- reviewing their device list.
-- * `last_seen_at` — refreshed every time the row authenticates a
-- request. Lets the revoke-device UI surface "last used 3 days
-- ago" so the user can tell which row corresponds to which
-- device.
-- * `revoked_at` — NULL means active; non-NULL stamps when the user
-- (or admin) revoked the row. Lookups treat any non-NULL value
-- as "this row is dead" without consulting the expiry; the
-- revoke gesture is intentionally one-way (a revoked device must
-- re-trust to come back online).
--
-- Indexing: a unique index on `device_token_hash` so collisions are
-- detectable at insert time (the token space is 256 bits of CSPRNG
-- entropy, so a collision is structurally impossible, but the
-- declaration documents the invariant). A separate index on
-- `(user_id, revoked_at)` so the revoke-device UI's list query is
-- a covering walk.
--
-- The bcrypt dependency reused here was added in v0.7.0 for OTC and
-- extended in v0.10.0 for passcodes; v0.11.0 needs no new dep.
--
-- The cookie shape: `rfc_device_trust` carries the raw token,
-- HttpOnly, Secure, SameSite=Lax, Max-Age=2592000 (30 days). It is
-- "essential" per the v0.13.0 cookie-consent banner (it is part of
-- authentication, not analytics), so it is set regardless of the
-- user's analytics / other-cookies choices.
CREATE TABLE device_trust (
id INTEGER PRIMARY KEY AUTOINCREMENT,
user_id INTEGER NOT NULL REFERENCES users(id) ON DELETE CASCADE,
device_token_hash TEXT NOT NULL,
created_at TEXT NOT NULL DEFAULT (datetime('now')),
expires_at TEXT NOT NULL,
user_agent TEXT NOT NULL DEFAULT '',
last_seen_at TEXT NOT NULL DEFAULT (datetime('now')),
revoked_at TEXT
);
CREATE UNIQUE INDEX idx_device_trust_token_hash ON device_trust (device_token_hash);
CREATE INDEX idx_device_trust_user ON device_trust (user_id, revoked_at);
+425
View File
@@ -0,0 +1,425 @@
"""End-to-end integration tests for v0.9.0's admin user-management page
and new-beta-request notifications (roadmap item #7, §6.1 / §15).
The release lands two halves of the same surface:
* **Admin notification on new beta request.** When a pending user
submits `POST /api/auth/me/beta-request`, every owner/admin
receives a `new_beta_request` notification (the §15 substrate
insert lands the row; the §15.4 email path dispatches subject to
the recipient's `email_admin_actionable` toggle).
* **Admin user-management surface** at `/admin/users`. The
`GET /api/admin/users` listing carries every user with their
permission_state, profile fields, sign-up reason, and decision
audit. The new `POST /api/admin/users/<id>/permission` endpoint
flips the column and writes a `permission_events` row.
The tests prove:
* The first beta-request submission fans a `new_beta_request`
row out to every admin/owner (and not to the requester
themselves). The row carries the captured profile in
`payload.extras`.
* Re-submitting the form from the same pending user doesn't
re-fan (we only notify on the row's first complete state).
* `GET /api/admin/users` carries the v0.9.0 columns
(permission_state, first/last/reason, decided_by).
* `POST /api/admin/users/<id>/permission` flips the state,
stamps decided_by/at, and writes a `permission_events` row.
* The endpoint refuses self-flip (422) and refuses non-admin
callers (403).
* The endpoint accepts only the three valid states (422 on
anything else).
"""
from __future__ import annotations
from test_propose_vertical import ( # noqa: F401 — fixtures land via import
FakeGitea,
app_with_fake_gitea,
provision_user_row,
sign_in_as,
tmp_env,
)
def _reset_outbound():
from app import email as email_mod
email_mod.reset_sent_envelopes()
def _outbound_otc_codes(to_address: str | None = None) -> list[str]:
from app import email as email_mod
out = []
for env in email_mod.sent_envelopes():
if env.get("kind") != "otc":
continue
if to_address is not None and env["to"] != to_address:
continue
for line in env["body"].splitlines():
tok = line.strip()
if tok.isdigit() and len(tok) == 6:
out.append(tok)
break
return out
def _provision_pending_user(client, email: str) -> int:
"""Sign in a fresh OTC user (lands `pending`) and return their user_id."""
from app import db
_reset_outbound()
client.post("/auth/otc/request", json={"email": email})
code = _outbound_otc_codes(email)[-1]
client.post("/auth/otc/verify", json={"email": email, "code": code})
row = db.conn().execute(
"SELECT id FROM users WHERE email = ? COLLATE NOCASE", (email,)
).fetchone()
return row["id"]
# ---------------------------------------------------------------------------
# Admin notification on beta-request submission
# ---------------------------------------------------------------------------
def test_beta_request_submission_notifies_every_admin(app_with_fake_gitea):
"""First-time submission of a beta-request fans a notification out
to every owner and admin. The requester themselves never receives
a row (filtered out by user_id even if they happened to be in the
admin set, which they aren't in practice — fresh OTC users are
`contributor`+`pending`)."""
from fastapi.testclient import TestClient
from app import db
app, _fake = app_with_fake_gitea
with TestClient(app) as client:
# Provision two admins and one owner so the fan-out has multiple
# targets. The OWNER_GITEA_LOGIN-derived ownership doesn't fire
# here (no OAuth round-trip in this path); we seed the role
# directly.
provision_user_row(user_id=10, login="ownerzero", role="owner")
provision_user_row(user_id=11, login="admin_one", role="admin")
provision_user_row(user_id=12, login="admin_two", role="admin")
provision_user_row(user_id=13, login="contrib_one", role="contributor")
# Sign in a fresh OTC user → permission_state='pending'.
requester_id = _provision_pending_user(client, "newbie@example.com")
# Capture-form submit.
r = client.post(
"/api/auth/me/beta-request",
json={
"first_name": "Newt",
"last_name": "Newcomer",
"beta_request_reason": "I want to write the Human RFC.",
},
)
assert r.status_code == 200, r.text
# Every owner + admin gets a `new_beta_request` notification.
# The contributor (id=13) does not. The requester (whoever id
# they got) does not.
rows = db.conn().execute(
"""
SELECT recipient_user_id, event_kind, actor_user_id, payload
FROM notifications
WHERE event_kind = 'new_beta_request'
"""
).fetchall()
recipients = sorted(r["recipient_user_id"] for r in rows)
assert recipients == [10, 11, 12], f"unexpected recipients: {recipients}"
# Actor is the requester (§15.9: never the bot).
for r in rows:
assert r["actor_user_id"] == requester_id
import json as _json
extras = _json.loads(r["payload"])
assert extras["requester_first_name"] == "Newt"
assert extras["requester_last_name"] == "Newcomer"
assert extras["requester_email"] == "newbie@example.com"
def test_beta_request_resubmit_does_not_re_notify(app_with_fake_gitea):
"""Once a user has completed the capture form, re-submitting it
(the endpoint is idempotent for pending users) must not re-fan a
fresh notification to every admin that would carpet-bomb the
inbox on every typo correction."""
from fastapi.testclient import TestClient
from app import db
app, _fake = app_with_fake_gitea
with TestClient(app) as client:
provision_user_row(user_id=20, login="adminzero", role="admin")
_provision_pending_user(client, "carpet@example.com")
body = {
"first_name": "Carpet",
"last_name": "Bomb",
"beta_request_reason": "first draft",
}
r1 = client.post("/api/auth/me/beta-request", json=body)
assert r1.status_code == 200
# Re-submit with edited reason — endpoint accepts (idempotent
# update), but the admin inbox stays at one row.
body2 = dict(body, beta_request_reason="cleaner final draft")
r2 = client.post("/api/auth/me/beta-request", json=body2)
assert r2.status_code == 200
rows = db.conn().execute(
"SELECT COUNT(*) AS n FROM notifications WHERE event_kind = 'new_beta_request'"
).fetchone()
assert rows["n"] == 1
def test_beta_request_notification_is_admin_actionable_category(app_with_fake_gitea):
"""The §15.4 category mapping must route `new_beta_request` to the
admin-actionable bucket so the email gate consults
`email_admin_actionable` (and skips for non-admin recipients).
"""
from app import email as email_mod
assert email_mod.category_for("new_beta_request", "structural") == "admin-actionable"
# ---------------------------------------------------------------------------
# /api/admin/users — listing carries the v0.9.0 columns
# ---------------------------------------------------------------------------
def test_admin_users_listing_carries_permission_columns(app_with_fake_gitea):
"""The Users tab consumes this shape — confirm every required
column is on the response."""
from fastapi.testclient import TestClient
from app import db
app, _fake = app_with_fake_gitea
with TestClient(app) as client:
# Seed an admin and a pending user with all the v0.8.0 columns
# populated. Direct-DB insert avoids the OTC dance (which would
# overwrite the cookie); the test above proves the capture
# pathway end-to-end and this one just exercises the listing
# surface's shape.
provision_user_row(user_id=30, login="ben", role="owner")
db.conn().execute(
"""
INSERT INTO users (id, gitea_id, gitea_login, email,
display_name, avatar_url, role,
permission_state, first_name, last_name,
beta_request_reason)
VALUES (31, NULL, NULL, 'pendinguser@example.com',
'pendinguser', '', 'contributor',
'pending', 'Penn', 'Ding', 'I want in.')
"""
)
sign_in_as(
client, user_id=30, gitea_login="ben",
display_name="Ben", role="owner",
)
r = client.get("/api/admin/users")
assert r.status_code == 200
items = r.json()["items"]
assert isinstance(items, list)
pending = next(
(i for i in items if i["email"] == "pendinguser@example.com"), None,
)
assert pending is not None
assert pending["permission_state"] == "pending"
assert pending["first_name"] == "Penn"
assert pending["last_name"] == "Ding"
assert pending["beta_request_reason"] == "I want in."
assert pending["permission_decided_at"] is None
assert pending["permission_decided_by_login"] is None
# Pending bucket is listed first (sort order).
assert items[0]["permission_state"] == "pending"
# ---------------------------------------------------------------------------
# /api/admin/users/<id>/permission — the flip endpoint
# ---------------------------------------------------------------------------
def test_permission_flip_grant_promotes_pending_to_granted(app_with_fake_gitea):
"""The end-to-end gesture: a fresh OTC user lands pending, an admin
flips them to granted via the endpoint, the row reflects the new
state + decided_by/at, and a `permission_events` audit row lands."""
from fastapi.testclient import TestClient
from app import db
app, _fake = app_with_fake_gitea
with TestClient(app) as client:
# Pending user.
pending_id = _provision_pending_user(client, "flip@example.com")
# Admin acting on them.
provision_user_row(user_id=40, login="adminflipper", role="admin")
sign_in_as(
client, user_id=40, gitea_login="adminflipper",
display_name="Admin Flipper", role="admin",
)
r = client.post(
f"/api/admin/users/{pending_id}/permission",
json={"state": "granted"},
)
assert r.status_code == 200, r.text
body = r.json()
assert body["permission_state"] == "granted"
assert body["changed"] is True
# Row reflects the new state + decision stamp.
row = db.conn().execute(
"SELECT permission_state, permission_decided_by, permission_decided_at "
"FROM users WHERE id = ?",
(pending_id,),
).fetchone()
assert row["permission_state"] == "granted"
assert row["permission_decided_by"] == 40
assert row["permission_decided_at"] is not None
# Audit row landed in permission_events.
events = db.conn().execute(
"""
SELECT actor_user_id, subject_user_id, event_kind
FROM permission_events
WHERE event_kind = 'permission_granted'
"""
).fetchall()
assert len(events) == 1
assert events[0]["actor_user_id"] == 40
assert events[0]["subject_user_id"] == pending_id
def test_permission_flip_revoke_promotes_granted_to_revoked(app_with_fake_gitea):
"""Revoke is the symmetric gesture. Used when an account earned a
grant then later lost it (§6.1 / `revoked` state)."""
from fastapi.testclient import TestClient
from app import db
app, _fake = app_with_fake_gitea
with TestClient(app) as client:
provision_user_row(user_id=50, login="goner", role="contributor")
# Default permission_state is 'granted' via the column default.
provision_user_row(user_id=51, login="adminrevoker", role="admin")
sign_in_as(
client, user_id=51, gitea_login="adminrevoker",
display_name="Admin Revoker", role="admin",
)
r = client.post(
"/api/admin/users/50/permission",
json={"state": "revoked"},
)
assert r.status_code == 200, r.text
row = db.conn().execute(
"SELECT permission_state FROM users WHERE id = 50"
).fetchone()
assert row["permission_state"] == "revoked"
events = db.conn().execute(
"SELECT event_kind FROM permission_events "
"WHERE event_kind = 'permission_revoked' AND subject_user_id = 50"
).fetchall()
assert len(events) == 1
def test_permission_flip_refuses_self(app_with_fake_gitea):
"""Symmetric to set_mute / set_role: an admin can't self-flip.
The state-change channel for one's own grant is somebody else's
hand."""
from fastapi.testclient import TestClient
app, _fake = app_with_fake_gitea
with TestClient(app) as client:
provision_user_row(user_id=60, login="selfflipper", role="admin")
sign_in_as(
client, user_id=60, gitea_login="selfflipper",
display_name="Self Flipper", role="admin",
)
r = client.post(
"/api/admin/users/60/permission",
json={"state": "revoked"},
)
assert r.status_code == 422
def test_permission_flip_refuses_non_admin(app_with_fake_gitea):
"""The endpoint is admin-only (§17 admin/* requires require_admin).
A contributor caller is refused 403; an anonymous caller 401."""
from fastapi.testclient import TestClient
app, _fake = app_with_fake_gitea
with TestClient(app) as client:
provision_user_row(user_id=70, login="target", role="contributor")
provision_user_row(user_id=71, login="contrib", role="contributor")
sign_in_as(
client, user_id=71, gitea_login="contrib",
display_name="Contrib", role="contributor",
)
r = client.post(
"/api/admin/users/70/permission",
json={"state": "granted"},
)
assert r.status_code == 403
client.cookies.clear()
r = client.post(
"/api/admin/users/70/permission",
json={"state": "granted"},
)
assert r.status_code == 401
def test_permission_flip_refuses_invalid_state(app_with_fake_gitea):
"""Pydantic regex pattern refuses anything outside the three
canonical states with 422."""
from fastapi.testclient import TestClient
app, _fake = app_with_fake_gitea
with TestClient(app) as client:
provision_user_row(user_id=80, login="targetx", role="contributor")
provision_user_row(user_id=81, login="adminx", role="admin")
sign_in_as(
client, user_id=81, gitea_login="adminx",
display_name="Admin X", role="admin",
)
r = client.post(
"/api/admin/users/80/permission",
json={"state": "banished"},
)
assert r.status_code == 422
def test_permission_flip_no_op_when_state_already_matches(app_with_fake_gitea):
"""An admin flipping a granted user to granted gets 200 with
`changed: false` no audit row, no decided_at update."""
from fastapi.testclient import TestClient
from app import db
app, _fake = app_with_fake_gitea
with TestClient(app) as client:
provision_user_row(user_id=90, login="alreadygranted", role="contributor")
provision_user_row(user_id=91, login="adminN", role="admin")
sign_in_as(
client, user_id=91, gitea_login="adminN",
display_name="Admin N", role="admin",
)
before_events = db.conn().execute(
"SELECT COUNT(*) AS n FROM permission_events"
).fetchone()["n"]
r = client.post(
"/api/admin/users/90/permission",
json={"state": "granted"},
)
assert r.status_code == 200
body = r.json()
assert body["changed"] is False
after_events = db.conn().execute(
"SELECT COUNT(*) AS n FROM permission_events"
).fetchone()["n"]
assert after_events == before_events
+390
View File
@@ -0,0 +1,390 @@
"""End-to-end integration tests for v0.8.0's open beta-access request
flow (§6.1 / §14.1, roadmap item #6).
The release replaces v0.3.0's `allowed_emails` allowlist as the
admission gate. Any valid email can sign in via the v0.7.0 OTC flow;
a fresh user lands in `permission_state='pending'` until an admin
grants access. The first-OTC flow captures first name, last name,
and a free-text "why I should be included in the beta" via a new
`POST /api/auth/me/beta-request` endpoint.
The tests prove:
* A fresh OTC user lands `permission_state='pending'` with empty
profile fields, and the verify-response carries `needs_profile=true`.
* `POST /api/auth/me/beta-request` populates the three fields and
leaves the row in `pending`.
* A pending user is refused write endpoints (representative
samples: propose RFC, post discussion thread). The refusal is
403 (not 401 they're authenticated, just not granted).
* An admin-grant flow promotes pending granted. v0.8.0 doesn't
ship an admin UI for this (deferred to item #7 / v0.9.0), so
the test flips the column directly via DB and asserts that
`require_contributor` now admits the user.
* A grandfathered user (existing row pre-migration, default
`permission_state='granted'`) is unaffected write endpoints
accept them.
* The `/auth/otc/request` endpoint accepts any email the
v0.7.0 allowlist gate is gone from this path. The `allowed_emails`
table stays in the schema; the admin UI from v0.3.0 continues to
manage it for the fast-path bypass deployments may use.
"""
from __future__ import annotations
import pytest
from test_propose_vertical import ( # noqa: F401 — fixtures land via import
FakeGitea,
app_with_fake_gitea,
provision_user_row,
sign_in_as,
tmp_env,
)
def _reset_outbound():
from app import email as email_mod
email_mod.reset_sent_envelopes()
def _outbound_otc_codes(to_address: str | None = None) -> list[str]:
"""Pluck the code line from every OTC envelope in the test buffer."""
from app import email as email_mod
out = []
for env in email_mod.sent_envelopes():
if env.get("kind") != "otc":
continue
if to_address is not None and env["to"] != to_address:
continue
for line in env["body"].splitlines():
tok = line.strip()
if tok.isdigit() and len(tok) == 6:
out.append(tok)
break
return out
# ---------------------------------------------------------------------------
# Fresh OTC sign-in lands pending with empty fields
# ---------------------------------------------------------------------------
def test_fresh_otc_user_lands_pending_with_empty_profile(app_with_fake_gitea):
from fastapi.testclient import TestClient
from app import db
app, _fake = app_with_fake_gitea
with TestClient(app) as client:
_reset_outbound()
# Request + verify the OTC.
r = client.post("/auth/otc/request", json={"email": "newcomer@example.com"})
assert r.status_code == 200, r.text
code = _outbound_otc_codes("newcomer@example.com")[-1]
r = client.post("/auth/otc/verify", json={"email": "newcomer@example.com", "code": code})
assert r.status_code == 200, r.text
body = r.json()
# The verify response carries the new fields v0.8.0 added.
assert body["needs_profile"] is True
assert body["user"]["permission_state"] == "pending"
# The row reflects the same: pending state, no profile yet.
row = db.conn().execute(
"SELECT permission_state, first_name, last_name, beta_request_reason FROM users WHERE email = ? COLLATE NOCASE",
("newcomer@example.com",),
).fetchone()
assert row is not None
assert row["permission_state"] == "pending"
assert row["first_name"] is None
assert row["last_name"] is None
assert row["beta_request_reason"] is None
# /api/auth/me surfaces the same shape.
me = client.get("/api/auth/me").json()
assert me["authenticated"] is True
assert me["user"]["permission_state"] == "pending"
assert me["user"]["needs_profile"] is True
assert me["user"]["first_name"] == ""
assert me["user"]["last_name"] == ""
assert me["user"]["beta_request_reason"] == ""
# ---------------------------------------------------------------------------
# beta-request endpoint captures the fields and leaves state pending
# ---------------------------------------------------------------------------
def test_beta_request_populates_fields_keeps_state_pending(app_with_fake_gitea):
from fastapi.testclient import TestClient
from app import db
app, _fake = app_with_fake_gitea
with TestClient(app) as client:
_reset_outbound()
# Sign in the fresh user via the full OTC flow.
client.post("/auth/otc/request", json={"email": "alice@example.com"})
code = _outbound_otc_codes("alice@example.com")[-1]
client.post("/auth/otc/verify", json={"email": "alice@example.com", "code": code})
# Submit the capture form.
r = client.post(
"/api/auth/me/beta-request",
json={
"first_name": "Alice",
"last_name": "Liddell",
"beta_request_reason": "I want to help write the RFCs.",
},
)
assert r.status_code == 200, r.text
# The row reflects the captured fields; state stays pending.
row = db.conn().execute(
"SELECT permission_state, first_name, last_name, beta_request_reason FROM users WHERE email = ? COLLATE NOCASE",
("alice@example.com",),
).fetchone()
assert row["permission_state"] == "pending"
assert row["first_name"] == "Alice"
assert row["last_name"] == "Liddell"
assert row["beta_request_reason"] == "I want to help write the RFCs."
# /api/auth/me now reports needs_profile=false (fields are set).
me = client.get("/api/auth/me").json()
assert me["user"]["permission_state"] == "pending"
assert me["user"]["needs_profile"] is False
assert me["user"]["first_name"] == "Alice"
def test_beta_request_refuses_anonymous(app_with_fake_gitea):
"""The endpoint requires authentication — an anonymous caller can't
file a request without first signing in via OTC."""
from fastapi.testclient import TestClient
app, _fake = app_with_fake_gitea
with TestClient(app) as client:
client.cookies.clear()
r = client.post(
"/api/auth/me/beta-request",
json={"first_name": "A", "last_name": "B", "beta_request_reason": "Hi"},
)
assert r.status_code == 401
def test_beta_request_refuses_granted_user(app_with_fake_gitea):
"""A grandfathered (already granted) user has no business filing a
beta request. The endpoint refuses with 409 so the client can
distinguish the failure from "we don't know you" (401)."""
from fastapi.testclient import TestClient
app, _fake = app_with_fake_gitea
with TestClient(app) as client:
provision_user_row(user_id=1, login="grandfathered", role="contributor")
sign_in_as(
client,
user_id=1,
gitea_login="grandfathered",
display_name="Grandfathered",
role="contributor",
)
r = client.post(
"/api/auth/me/beta-request",
json={"first_name": "G", "last_name": "F", "beta_request_reason": "x"},
)
assert r.status_code == 409
# ---------------------------------------------------------------------------
# Pending user is refused write endpoints; admin grant promotes them
# ---------------------------------------------------------------------------
def test_pending_user_is_refused_write_endpoints(app_with_fake_gitea):
"""A pending user can read everything anonymous can read, but every
write-shaped endpoint refuses with 403. The refusal shape mirrors
the v0.6.0 / item #4 audit's anon-401 — both are "no contributor
capability"; pending is the authenticated-but-ungranted variant.
Representative samples: propose RFC, post discussion thread.
"""
from fastapi.testclient import TestClient
app, _fake = app_with_fake_gitea
with TestClient(app) as client:
_reset_outbound()
# Sign in via fresh OTC — lands pending.
client.post("/auth/otc/request", json={"email": "pending@example.com"})
code = _outbound_otc_codes("pending@example.com")[-1]
client.post("/auth/otc/verify", json={"email": "pending@example.com", "code": code})
# Reads work — every anonymous surface stays reachable.
assert client.get("/api/health").status_code == 200
assert client.get("/api/rfcs").status_code == 200
assert client.get("/api/philosophy").status_code == 200
# Propose — write-shaped, refused with 403.
r = client.post(
"/api/rfcs/propose",
json={"title": "T", "slug": "t", "pitch": "p", "tags": []},
)
assert r.status_code == 403
# The error body mentions the review state so a UI surface can
# render the right message — but the test asserts only on the
# status code (the body shape is the FastAPI default detail).
def test_admin_grant_promotes_pending_to_granted(app_with_fake_gitea):
"""v0.8.0 doesn't ship an admin UI for this — it's deferred to
item #7 / v0.9.0. For this release, an admin gesture is an
`UPDATE users SET permission_state='granted' WHERE email=?`. The
test flips the column directly via DB and asserts the
`require_contributor` gate now admits the user.
"""
from fastapi.testclient import TestClient
from app import db
app, _fake = app_with_fake_gitea
with TestClient(app) as client:
_reset_outbound()
# Sign in a fresh OTC user — lands pending.
client.post("/auth/otc/request", json={"email": "promoted@example.com"})
code = _outbound_otc_codes("promoted@example.com")[-1]
client.post("/auth/otc/verify", json={"email": "promoted@example.com", "code": code})
# Before the grant: propose refused with 403.
r = client.post(
"/api/rfcs/propose",
json={"title": "T", "slug": "t-pre", "pitch": "p", "tags": []},
)
assert r.status_code == 403
# The admin gesture (v0.8.0 shape — direct UPDATE; v0.9.0 will
# ship a UI). The test stamps `permission_decided_by` and
# `permission_decided_at` as the v0.9.0 admin UI will, so the
# column population exercises the schema slot. user_id=99 is
# a placeholder admin row — provision it so the FK resolves.
provision_user_row(user_id=99, login="adminuser", role="admin")
db.conn().execute(
"""
UPDATE users
SET permission_state = 'granted',
permission_decided_by = 99,
permission_decided_at = datetime('now')
WHERE email = ?
""",
("promoted@example.com",),
)
# The next request reads the fresh column from the DB. The
# propose endpoint reaches the route body now (it then refuses
# for a different reason — the slug 't-prop' will fail
# the slug-format check or hit a mock-gitea path — but the
# status code is _not_ 403/401, which is the v0.8.0 assertion).
r = client.post(
"/api/rfcs/propose",
json={"title": "Title", "slug": "tprop", "pitch": "Pitch text.", "tags": []},
)
assert r.status_code != 403, r.text
assert r.status_code != 401, r.text
def test_grandfathered_user_is_unaffected_by_migration(app_with_fake_gitea):
"""An existing `users` row at migration time has
`permission_state='granted'` via the column default. The
grandfathered user passes write endpoints without filing a
beta request and without the admin UI. v0.6.0 (anon-write
audit) is the v0.6.0 contract; v0.8.0 widens the gate but
does not break this case.
"""
from fastapi.testclient import TestClient
from app import db
app, _fake = app_with_fake_gitea
with TestClient(app) as client:
provision_user_row(user_id=5, login="oldhand", role="contributor")
# provision_user_row uses INSERT OR REPLACE INTO users with
# the column list it knows; permission_state is not in that
# list, so it picks up the column default ('granted') on
# insert. Confirm directly.
row = db.conn().execute(
"SELECT permission_state FROM users WHERE id = 5"
).fetchone()
assert row["permission_state"] == "granted"
sign_in_as(
client,
user_id=5,
gitea_login="oldhand",
display_name="Old Hand",
role="contributor",
)
# Propose is write-shaped; the call should not refuse on
# the permission_state gate. (Subsequent failure modes —
# e.g. mock-gitea wiring — are not the v0.8.0 concern; this
# test asserts on the gate, not the propose body's success.)
r = client.post(
"/api/rfcs/propose",
json={"title": "Title", "slug": "gf-slug", "pitch": "Pitch.", "tags": []},
)
assert r.status_code != 403, r.text
assert r.status_code != 401, r.text
# ---------------------------------------------------------------------------
# /auth/otc/request accepts any email — the v0.7.0 allowlist gate is gone
# ---------------------------------------------------------------------------
def test_otc_request_accepts_any_email_regardless_of_allowlist(app_with_fake_gitea):
"""v0.7.0 silently dropped OTC requests for emails not on the
`allowed_emails` table. v0.8.0 reverses this: the request
endpoint sends a code to any valid email; admission gates at
`permission_state` post-verify instead. The `allowed_emails`
table stays in the schema as a fast-path bypass for
deployments that want to pre-mark known-good emails (the v0.9.0
admin user-management page will collapse the two surfaces).
"""
from fastapi.testclient import TestClient
from app import db
app, _fake = app_with_fake_gitea
with TestClient(app) as client:
_reset_outbound()
# Populate the allowlist with one specific email so the v0.7.0
# gate would have engaged. v0.8.0 ignores it for the request
# path.
db.conn().execute("INSERT INTO allowed_emails (email) VALUES (?)", ("known@example.com",))
# An email NOT on the allowlist still gets a code under v0.8.0.
r = client.post("/auth/otc/request", json={"email": "stranger@example.com"})
assert r.status_code == 200
codes = _outbound_otc_codes("stranger@example.com")
assert len(codes) == 1, "OTC code must be sent regardless of allowlist state"
# The row is there and the user can complete sign-in (and will
# land in 'pending' per the other tests).
row = db.conn().execute(
"SELECT 1 FROM otc_codes WHERE email = ?",
("stranger@example.com",),
).fetchone()
assert row is not None
def test_allowlist_table_still_present_in_schema(app_with_fake_gitea):
"""The schema migration leaves the `allowed_emails` table in
place the admin UI from v0.3.0 still manages it for the
fast-path bypass deployments may use. This is a regression net
for "did the v0.8.0 cleanup accidentally drop the table"."""
from fastapi.testclient import TestClient
from app import db
app, _fake = app_with_fake_gitea
with TestClient(app):
# The table accepts inserts (i.e. it exists) — no schema check
# gymnastics needed.
db.conn().execute("INSERT INTO allowed_emails (email) VALUES (?)", ("kept@example.com",))
row = db.conn().execute(
"SELECT email FROM allowed_emails WHERE email = ?",
("kept@example.com",),
).fetchone()
assert row is not None
+494
View File
@@ -0,0 +1,494 @@
"""End-to-end integration tests for the v0.11.0 trust-device vertical
(§6.2, roadmap item #9).
After a successful OTC or passcode sign-in with `trust_device=true`
on the body, the server mints a fresh `device_trust` row and sets the
`rfc_device_trust` cookie. On a subsequent visit, the cookie carries
a session re-established by `POST /auth/device-trust/start`. The
tests below prove:
* `trust_device=false` (default, including omitted) on OTC verify
does NOT set the device-trust cookie and does NOT insert a row.
* `trust_device=true` on OTC verify DOES set the cookie (HttpOnly +
Secure + SameSite=Lax + 30-day Max-Age) and DOES insert a row.
The row's hash is NOT the raw token; only the hash lives in the
database.
* Same shape for passcode verify.
* On a returning visit with the cookie, `POST /auth/device-trust/start`
re-establishes the session `GET /api/auth/me` reads the right
user without an OTC roundtrip.
* `last_seen_at` refreshes on a successful lookup.
* `POST /auth/device-trust/start` with no cookie returns 401.
* `POST /auth/device-trust/start` with a forged / unknown cookie
returns 401 + clears the cookie.
* A revoked row refuses the cookie (401) and clears it.
* An expired row refuses the cookie (401) and clears it.
* `GET /api/auth/me/devices` lists the user's active rows.
* `DELETE /api/auth/me/devices/{id}` revokes a single row.
* `DELETE /api/auth/me/devices/{id}` for another user's row reads 404.
* `DELETE /api/auth/me/devices` revokes every active row.
* Constant-time path: bcrypt.checkpw guards lookup; the raw token
is never written to logs or to the DB.
The fakes from `test_propose_vertical` give us a working app harness.
The OTC envelope buffer from `test_otc_vertical` is reused for the
OTC roundtrips this suite needs.
"""
from __future__ import annotations
import pytest
from test_propose_vertical import ( # noqa: F401
FakeGitea,
app_with_fake_gitea,
provision_user_row,
tmp_env,
)
# ---------------------------------------------------------------------------
# Helpers — mirror the OTC suite's outbound-buffer helpers.
# ---------------------------------------------------------------------------
COOKIE_NAME = "rfc_device_trust"
# The device-trust cookie is set with Secure=True, which httpx (the
# TestClient's underlying transport) will only return on an https
# scheme. We use a `base_url="https://testserver"` so the cookie
# roundtrips faithfully — that mirrors how production deployments
# serve the framework (per the v0.11.0 upgrade-step requiring HTTPS).
HTTPS_BASE = "https://testserver"
def _reset_outbound():
from app import email as email_mod
email_mod.reset_sent_envelopes()
def _outbound_otc_codes(to_address: str | None = None) -> list[str]:
from app import email as email_mod
out = []
for env in email_mod.sent_envelopes():
if env.get("kind") != "otc":
continue
if to_address is not None and env["to"] != to_address:
continue
for line in env["body"].splitlines():
tok = line.strip()
if tok.isdigit() and len(tok) == 6:
out.append(tok)
break
return out
def _sign_in_via_otc(client, email: str, *, trust_device: bool = False) -> None:
r = client.post("/auth/otc/request", json={"email": email})
assert r.status_code == 200, r.text
code = _outbound_otc_codes(email)[-1]
body = {"email": email, "code": code, "trust_device": trust_device}
r = client.post("/auth/otc/verify", json=body)
assert r.status_code == 200, r.text
def _device_rows_for_email(email: str) -> list[dict]:
from app import db
rows = db.conn().execute(
"""
SELECT dt.*
FROM device_trust dt
JOIN users u ON u.id = dt.user_id
WHERE u.email = ? COLLATE NOCASE
ORDER BY dt.id
""",
(email,),
).fetchall()
return [dict(r) for r in rows]
# ---------------------------------------------------------------------------
# trust_device flag controls cookie issuance
# ---------------------------------------------------------------------------
def test_otc_verify_without_trust_device_does_not_issue_cookie(app_with_fake_gitea):
from fastapi.testclient import TestClient
app, _fake = app_with_fake_gitea
with TestClient(app, base_url=HTTPS_BASE) as client:
_reset_outbound()
_sign_in_via_otc(client, "alice@example.com", trust_device=False)
# No cookie set on the response.
assert COOKIE_NAME not in {c.name for c in client.cookies.jar}
# No row inserted.
assert _device_rows_for_email("alice@example.com") == []
def test_otc_verify_with_trust_device_issues_cookie_and_row(app_with_fake_gitea):
from fastapi.testclient import TestClient
app, _fake = app_with_fake_gitea
with TestClient(app, base_url=HTTPS_BASE) as client:
_reset_outbound()
r = client.post("/auth/otc/request", json={"email": "alice@example.com"})
assert r.status_code == 200, r.text
code = _outbound_otc_codes("alice@example.com")[-1]
r = client.post(
"/auth/otc/verify",
json={"email": "alice@example.com", "code": code, "trust_device": True},
headers={"User-Agent": "Mozilla/5.0 (TestBrowser)"},
)
assert r.status_code == 200, r.text
# Cookie present on the response.
set_cookie = r.headers.get("set-cookie", "")
assert COOKIE_NAME in set_cookie
# Cookie attribute set asserts the spec'd shape. Starlette emits
# the attribute names case-insensitively (`samesite=lax`,
# `httponly`); we normalize when asserting.
lower = set_cookie.lower()
assert "httponly" in lower
assert "secure" in lower
assert "samesite=lax" in lower
assert "max-age=" in lower
# Row inserted; hash is not the raw token.
rows = _device_rows_for_email("alice@example.com")
assert len(rows) == 1
row = rows[0]
assert row["revoked_at"] is None
assert row["user_agent"] == "Mozilla/5.0 (TestBrowser)"
cookie_token = client.cookies.get(COOKIE_NAME)
assert cookie_token
assert cookie_token != row["device_token_hash"]
# bcrypt hash shape (starts with $2)
assert row["device_token_hash"].startswith("$2")
def test_passcode_verify_with_trust_device_issues_cookie(app_with_fake_gitea):
from fastapi.testclient import TestClient
app, _fake = app_with_fake_gitea
with TestClient(app, base_url=HTTPS_BASE) as client:
_reset_outbound()
_sign_in_via_otc(client, "alice@example.com")
# Set a passcode.
r = client.post("/auth/passcode/set", json={"passcode": "secret123"})
assert r.status_code == 200, r.text
# Sign out so the passcode verify path is the active sign-in.
client.cookies.clear()
# Passcode verify with trust_device=true issues a row.
r = client.post(
"/auth/passcode/verify",
json={"email": "alice@example.com", "passcode": "secret123", "trust_device": True},
headers={"User-Agent": "Test/Phone"},
)
assert r.status_code == 200, r.text
set_cookie = r.headers.get("set-cookie", "")
assert COOKIE_NAME in set_cookie
rows = _device_rows_for_email("alice@example.com")
assert len(rows) == 1
assert rows[0]["user_agent"] == "Test/Phone"
# ---------------------------------------------------------------------------
# /auth/device-trust/start
# ---------------------------------------------------------------------------
def test_device_trust_start_with_no_cookie_returns_401(app_with_fake_gitea):
from fastapi.testclient import TestClient
app, _fake = app_with_fake_gitea
with TestClient(app, base_url=HTTPS_BASE) as client:
r = client.post("/auth/device-trust/start")
assert r.status_code == 401
def test_device_trust_start_with_valid_cookie_establishes_session(app_with_fake_gitea):
from fastapi.testclient import TestClient
app, _fake = app_with_fake_gitea
with TestClient(app, base_url=HTTPS_BASE) as client:
_reset_outbound()
# Trust the device.
r = client.post("/auth/otc/request", json={"email": "alice@example.com"})
assert r.status_code == 200, r.text
code = _outbound_otc_codes("alice@example.com")[-1]
r = client.post(
"/auth/otc/verify",
json={"email": "alice@example.com", "code": code, "trust_device": True},
)
assert r.status_code == 200, r.text
trust_cookie = client.cookies.get(COOKIE_NAME)
assert trust_cookie
# Clear the session cookie so only the device-trust cookie is in
# play. We keep `rfc_device_trust` and drop `rfc_session`.
for cookie in list(client.cookies.jar):
if cookie.name != COOKIE_NAME:
client.cookies.jar.clear(cookie.domain, cookie.path, cookie.name)
# The session cookie is gone — /api/auth/me reads anonymous.
me = client.get("/api/auth/me").json()
assert me["authenticated"] is False
# Hit the trust-start endpoint; the cookie re-establishes the session.
r = client.post("/auth/device-trust/start")
assert r.status_code == 200, r.text
assert r.json()["user"]["email"] == "alice@example.com"
# /api/auth/me now reads authenticated.
me = client.get("/api/auth/me").json()
assert me["authenticated"] is True
assert me["user"]["email"] == "alice@example.com"
def test_device_trust_start_refreshes_last_seen_at(app_with_fake_gitea):
from fastapi.testclient import TestClient
from app import db
app, _fake = app_with_fake_gitea
with TestClient(app, base_url=HTTPS_BASE) as client:
_reset_outbound()
r = client.post("/auth/otc/request", json={"email": "alice@example.com"})
assert r.status_code == 200, r.text
code = _outbound_otc_codes("alice@example.com")[-1]
r = client.post(
"/auth/otc/verify",
json={"email": "alice@example.com", "code": code, "trust_device": True},
)
assert r.status_code == 200, r.text
# Force the existing row's last_seen_at into the past so we can
# assert the refresh moved it forward.
db.conn().execute(
"""
UPDATE device_trust
SET last_seen_at = datetime('now', '-7 days')
"""
)
# Hit the start endpoint.
r = client.post("/auth/device-trust/start")
assert r.status_code == 200, r.text
# last_seen_at is now recent (within the last minute).
row = db.conn().execute(
"SELECT last_seen_at, datetime('now') >= datetime(last_seen_at, '-1 minute') AS fresh FROM device_trust LIMIT 1"
).fetchone()
assert row["fresh"] == 1
def test_device_trust_start_with_revoked_row_refuses_and_clears(app_with_fake_gitea):
from fastapi.testclient import TestClient
from app import db
app, _fake = app_with_fake_gitea
with TestClient(app, base_url=HTTPS_BASE) as client:
_reset_outbound()
r = client.post("/auth/otc/request", json={"email": "alice@example.com"})
assert r.status_code == 200, r.text
code = _outbound_otc_codes("alice@example.com")[-1]
r = client.post(
"/auth/otc/verify",
json={"email": "alice@example.com", "code": code, "trust_device": True},
)
assert r.status_code == 200, r.text
# Revoke the row out-of-band.
db.conn().execute(
"UPDATE device_trust SET revoked_at = datetime('now')"
)
# Now the start endpoint refuses + clears the cookie.
r = client.post("/auth/device-trust/start")
assert r.status_code == 401
# The cookie is cleared via a Set-Cookie header with Max-Age=0
# (Starlette's `delete_cookie` shape).
set_cookie = r.headers.get("set-cookie", "")
assert COOKIE_NAME in set_cookie
assert "Max-Age=0" in set_cookie or 'expires=Thu, 01 Jan 1970' in set_cookie.lower().replace("expires=thu", "expires=Thu")
def test_device_trust_start_with_expired_row_refuses_and_clears(app_with_fake_gitea):
from fastapi.testclient import TestClient
from app import db
app, _fake = app_with_fake_gitea
with TestClient(app, base_url=HTTPS_BASE) as client:
_reset_outbound()
r = client.post("/auth/otc/request", json={"email": "alice@example.com"})
assert r.status_code == 200, r.text
code = _outbound_otc_codes("alice@example.com")[-1]
r = client.post(
"/auth/otc/verify",
json={"email": "alice@example.com", "code": code, "trust_device": True},
)
assert r.status_code == 200, r.text
# Backdate the expiry into the past.
db.conn().execute(
"UPDATE device_trust SET expires_at = datetime('now', '-1 day')"
)
r = client.post("/auth/device-trust/start")
assert r.status_code == 401
set_cookie = r.headers.get("set-cookie", "")
assert COOKIE_NAME in set_cookie
def test_device_trust_start_with_forged_cookie_refuses_and_clears(app_with_fake_gitea):
from fastapi.testclient import TestClient
app, _fake = app_with_fake_gitea
with TestClient(app, base_url=HTTPS_BASE) as client:
# No real row; just paste a cookie value.
client.cookies.set(COOKIE_NAME, "definitely-not-a-real-token-value-xxx")
r = client.post("/auth/device-trust/start")
assert r.status_code == 401
set_cookie = r.headers.get("set-cookie", "")
assert COOKIE_NAME in set_cookie
# ---------------------------------------------------------------------------
# /api/auth/me/devices — list + revoke
# ---------------------------------------------------------------------------
def test_list_devices_requires_session(app_with_fake_gitea):
from fastapi.testclient import TestClient
app, _fake = app_with_fake_gitea
with TestClient(app, base_url=HTTPS_BASE) as client:
r = client.get("/api/auth/me/devices")
assert r.status_code == 401
def test_list_devices_returns_active_rows_only(app_with_fake_gitea):
from fastapi.testclient import TestClient
from app import db
app, _fake = app_with_fake_gitea
with TestClient(app, base_url=HTTPS_BASE) as client:
_reset_outbound()
_sign_in_via_otc(client, "alice@example.com", trust_device=True)
# Add a second trusted device by re-running the verify flow.
# OTC has a per-email cooldown, so drop the cooldown rather
# than waiting it out.
db.conn().execute("UPDATE otc_codes SET consumed_at = datetime('now', '-1 hour'), created_at = datetime('now', '-1 hour')")
r = client.post("/auth/otc/request", json={"email": "alice@example.com"})
assert r.status_code == 200, r.text
code = _outbound_otc_codes("alice@example.com")[-1]
r = client.post(
"/auth/otc/verify",
json={"email": "alice@example.com", "code": code, "trust_device": True},
headers={"User-Agent": "Test/Tablet"},
)
assert r.status_code == 200, r.text
# Revoke one row directly.
db.conn().execute(
"UPDATE device_trust SET revoked_at = datetime('now') WHERE id = 1"
)
# /api/auth/me/devices returns only the un-revoked one.
r = client.get("/api/auth/me/devices")
assert r.status_code == 200, r.text
items = r.json()["items"]
assert len(items) == 1
assert items[0]["user_agent"] == "Test/Tablet"
def test_revoke_single_device_kills_the_row(app_with_fake_gitea):
from fastapi.testclient import TestClient
from app import db
app, _fake = app_with_fake_gitea
with TestClient(app, base_url=HTTPS_BASE) as client:
_reset_outbound()
_sign_in_via_otc(client, "alice@example.com", trust_device=True)
r = client.get("/api/auth/me/devices")
assert r.status_code == 200, r.text
items = r.json()["items"]
assert len(items) == 1
device_id = items[0]["id"]
# Revoke it.
r = client.delete(f"/api/auth/me/devices/{device_id}")
assert r.status_code == 200, r.text
# List is empty.
r = client.get("/api/auth/me/devices")
assert r.json()["items"] == []
# The row in the table has revoked_at populated.
row = db.conn().execute(
"SELECT revoked_at FROM device_trust WHERE id = ?", (device_id,)
).fetchone()
assert row["revoked_at"] is not None
def test_revoke_other_users_device_reads_404(app_with_fake_gitea):
from fastapi.testclient import TestClient
from app import db
app, _fake = app_with_fake_gitea
with TestClient(app, base_url=HTTPS_BASE) as client:
_reset_outbound()
# Alice trusts a device.
_sign_in_via_otc(client, "alice@example.com", trust_device=True)
alice_device_id = client.get("/api/auth/me/devices").json()["items"][0]["id"]
# Bob signs in (without a trusted device of his own).
client.cookies.clear()
db.conn().execute("UPDATE otc_codes SET consumed_at = datetime('now', '-1 hour'), created_at = datetime('now', '-1 hour')")
_sign_in_via_otc(client, "bob@example.com", trust_device=False)
# Bob tries to revoke Alice's row by id.
r = client.delete(f"/api/auth/me/devices/{alice_device_id}")
assert r.status_code == 404
# Alice's row is still active.
row = db.conn().execute(
"SELECT revoked_at FROM device_trust WHERE id = ?", (alice_device_id,)
).fetchone()
assert row["revoked_at"] is None
def test_revoke_all_devices_kills_every_active_row(app_with_fake_gitea):
from fastapi.testclient import TestClient
from app import db
app, _fake = app_with_fake_gitea
with TestClient(app, base_url=HTTPS_BASE) as client:
_reset_outbound()
_sign_in_via_otc(client, "alice@example.com", trust_device=True)
# Add a second device.
db.conn().execute("UPDATE otc_codes SET consumed_at = datetime('now', '-1 hour'), created_at = datetime('now', '-1 hour')")
r = client.post("/auth/otc/request", json={"email": "alice@example.com"})
assert r.status_code == 200, r.text
code = _outbound_otc_codes("alice@example.com")[-1]
r = client.post(
"/auth/otc/verify",
json={"email": "alice@example.com", "code": code, "trust_device": True},
)
assert r.status_code == 200, r.text
# Two active rows.
assert len(client.get("/api/auth/me/devices").json()["items"]) == 2
# Revoke all.
r = client.delete("/api/auth/me/devices")
assert r.status_code == 200, r.text
assert r.json()["revoked"] == 2
# List is empty.
assert client.get("/api/auth/me/devices").json()["items"] == []
+30 -14
View File
@@ -13,9 +13,13 @@ sign-in path. The tests prove:
* Expired codes refuse with 400.
* Already-consumed codes refuse with 400 on re-use.
* Wrong codes refuse with 400.
* Allowlist gate: when `allowed_emails` is populated and the email
isn't on it, the response is still 202 (no leak), but no email
lands in the outbound buffer and verify finds no matching code.
* Allowlist gate (v0.8.0 update): v0.7.0 silently dropped requests
for emails not on `allowed_emails`. v0.8.0 (item #6) removed
that gate from the request path; the admission gate is now
`permission_state` on the freshly-provisioned `users` row,
asserted in test_beta_access_vertical.py. The tests below
confirm v0.8.0's open-request shape for both on-list and
off-list emails.
* Migration link: an existing OAuth-era user (with a `users.email`
row) is linked by email on first OTC sign-in `gitea_id` is
preserved.
@@ -201,33 +205,45 @@ def test_otc_request_cooldown_is_per_email_not_global(app_with_fake_gitea):
# ---------------------------------------------------------------------------
# Allowlist gate
# Allowlist gate — v0.8.0 update
#
# v0.7.0 gated the OTC request endpoint on the `allowed_emails` table:
# emails not on the list got a silent drop (still 202, but no code).
# v0.8.0 (roadmap item #6) reverses this: the request endpoint
# accepts any valid email and sends a code. The admission gate moves
# to `permission_state` on the freshly-provisioned `users` row,
# which the next-tier tests in test_beta_access_vertical.py cover.
# The `allowed_emails` table stays in the schema as a fast-path
# bypass for admin convenience.
# ---------------------------------------------------------------------------
def test_otc_request_silently_drops_when_email_not_on_allowlist(app_with_fake_gitea):
def test_otc_request_admits_emails_regardless_of_allowlist_population(app_with_fake_gitea):
"""v0.8.0: the OTC request path no longer consults `allowed_emails`.
Whether the allowlist is empty or populated, every valid email
receives a code; admission gates at `permission_state` post-verify.
"""
from fastapi.testclient import TestClient
from app import db
app, _fake = app_with_fake_gitea
with TestClient(app) as client:
_reset_outbound()
# Populate the allowlist so the gate turns on.
# Populate the allowlist with one specific email; the v0.7.0
# gate would have engaged here.
db.conn().execute("INSERT INTO allowed_emails (email) VALUES (?)", ("invited@example.com",))
# The not-on-list email still gets a code under v0.8.0.
r = client.post("/auth/otc/request", json={"email": "stranger@example.com"})
# Still 202 — the allowlist's state is not leaked to callers.
assert r.status_code == 200
# But no email was sent, and no row landed in otc_codes.
assert _outbound_otc_codes("stranger@example.com") == []
row = db.conn().execute(
"SELECT 1 FROM otc_codes WHERE email = ?",
("stranger@example.com",),
).fetchone()
assert row is None
assert len(_outbound_otc_codes("stranger@example.com")) == 1
def test_otc_request_admits_allowlisted_email(app_with_fake_gitea):
"""v0.8.0: still works for emails that happen to be on the legacy
allowlist the table is no longer consulted at request time but
populated rows are admitted alongside everyone else (since the
gate is now open at the request surface)."""
from fastapi.testclient import TestClient
from app import db
+220
View File
@@ -0,0 +1,220 @@
"""End-to-end integration tests for the v0.12.0 CloudFlare Turnstile
gate on `/auth/otc/request` (§6.2 / roadmap item #10).
The release gates the OTC request endpoint behind a one-step
browser-side Turnstile challenge before the bcrypt hash + SMTP send.
The tests prove:
* Happy path: with the secret set, a valid token admits the request
and the OTC envelope lands.
* Failure path: with the secret set, a token siteverify rejects
refuses the request with 400 and produces no envelope.
* Missing-token: with the secret set, a request without a token
refuses with 400.
* Missing-secret-soft: with the secret unset AND
`TURNSTILE_REQUIRED=false` (the v0.12.0 default), the request
admits this is the dev / "operator hasn't wired it yet" path.
* Missing-secret-hard: with the secret unset AND
`TURNSTILE_REQUIRED=true`, the request refuses with 500
"auth misconfigured" the production fail-closed path once
the operator has flipped the policy.
The Turnstile siteverify call is mocked at the `httpx.post` boundary
inside `app.turnstile` so no real keys are needed and no real
CloudFlare call is made. The Gitea fakes from `test_propose_vertical`
remain in scope so the rest of the app boots cleanly.
"""
from __future__ import annotations
from types import SimpleNamespace
import pytest
from test_propose_vertical import ( # noqa: F401
FakeGitea,
app_with_fake_gitea,
tmp_env,
)
def _reset_outbound():
from app import email as email_mod
email_mod.reset_sent_envelopes()
def _outbound_otc_envelopes(to_address: str | None = None) -> list[dict]:
from app import email as email_mod
out = []
for env in email_mod.sent_envelopes():
if env.get("kind") != "otc":
continue
if to_address is not None and env["to"] != to_address:
continue
out.append(env)
return out
def _patch_siteverify(monkeypatch, *, success: bool, error_codes: list[str] | None = None):
"""Replace `httpx.post` inside `app.turnstile` with a stub that
returns the requested success shape. The stub does not touch the
real CloudFlare endpoint and never sees a real secret.
"""
captured = {}
def fake_post(url, *, data=None, timeout=None, **kwargs):
captured["url"] = url
captured["data"] = data
body = {"success": bool(success)}
if error_codes is not None:
body["error-codes"] = error_codes
return SimpleNamespace(json=lambda: body)
from app import turnstile as turnstile_mod
monkeypatch.setattr(turnstile_mod.httpx, "post", fake_post)
return captured
# ---------------------------------------------------------------------------
# Happy path: secret set, token valid → admit + OTC envelope lands
# ---------------------------------------------------------------------------
def test_otc_request_admits_when_turnstile_token_is_valid(app_with_fake_gitea, monkeypatch):
from fastapi.testclient import TestClient
monkeypatch.setenv("CLOUDFLARE_TURNSTILE_SECRET", "test-secret-not-real")
monkeypatch.setenv("TURNSTILE_REQUIRED", "true")
captured = _patch_siteverify(monkeypatch, success=True)
app, _fake = app_with_fake_gitea
with TestClient(app) as client:
_reset_outbound()
r = client.post(
"/auth/otc/request",
json={"email": "alice@example.com", "turnstile_token": "fake-token-abc"},
)
assert r.status_code == 200, r.text
# The siteverify call was made with the secret + the token we sent.
assert captured["data"]["secret"] == "test-secret-not-real"
assert captured["data"]["response"] == "fake-token-abc"
# And the OTC dispatch ran — exactly one envelope to the address.
envs = _outbound_otc_envelopes("alice@example.com")
assert len(envs) == 1
# ---------------------------------------------------------------------------
# Failure path: secret set, siteverify says success=false → 400 + no envelope
# ---------------------------------------------------------------------------
def test_otc_request_refuses_when_turnstile_siteverify_fails(app_with_fake_gitea, monkeypatch):
from fastapi.testclient import TestClient
monkeypatch.setenv("CLOUDFLARE_TURNSTILE_SECRET", "test-secret-not-real")
monkeypatch.setenv("TURNSTILE_REQUIRED", "true")
_patch_siteverify(monkeypatch, success=False, error_codes=["invalid-input-response"])
app, _fake = app_with_fake_gitea
with TestClient(app) as client:
_reset_outbound()
r = client.post(
"/auth/otc/request",
json={"email": "alice@example.com", "turnstile_token": "fake-bad-token"},
)
assert r.status_code == 400, r.text
# The OTC bcrypt + SMTP path did not run — no envelope was buffered.
assert _outbound_otc_envelopes("alice@example.com") == []
# ---------------------------------------------------------------------------
# Missing-token: secret set, no token → 400 + no envelope
# ---------------------------------------------------------------------------
def test_otc_request_refuses_when_turnstile_token_is_missing(app_with_fake_gitea, monkeypatch):
from fastapi.testclient import TestClient
monkeypatch.setenv("CLOUDFLARE_TURNSTILE_SECRET", "test-secret-not-real")
monkeypatch.setenv("TURNSTILE_REQUIRED", "true")
# Even though we patch httpx.post, the missing-token check fires
# before the siteverify call — so the patch is here only as a
# safety net in case the implementation regresses to making the
# network call anyway.
_patch_siteverify(monkeypatch, success=False)
app, _fake = app_with_fake_gitea
with TestClient(app) as client:
_reset_outbound()
r = client.post(
"/auth/otc/request",
json={"email": "alice@example.com"}, # no turnstile_token field at all
)
assert r.status_code == 400, r.text
assert _outbound_otc_envelopes("alice@example.com") == []
# ---------------------------------------------------------------------------
# Missing-secret-soft: no secret, TURNSTILE_REQUIRED=false (default) → admit
# ---------------------------------------------------------------------------
def test_otc_request_admits_when_secret_unset_and_not_required(app_with_fake_gitea, monkeypatch):
"""v0.12.0 default: the operator has not yet wired the Turnstile
secret and has not enabled `TURNSTILE_REQUIRED`. The gate stays
open this is the dev / test / pre-rollout path. Once the
operator confirms the secret is in place and flips
`TURNSTILE_REQUIRED=true`, missing-secret becomes fail-closed
(covered in test_otc_request_refuses_when_required_but_secret_unset).
"""
from fastapi.testclient import TestClient
monkeypatch.delenv("CLOUDFLARE_TURNSTILE_SECRET", raising=False)
monkeypatch.delenv("TURNSTILE_REQUIRED", raising=False)
# The httpx.post inside turnstile must not be called in this path —
# patch it to a sentinel that explodes if it ever runs.
from app import turnstile as turnstile_mod
def must_not_be_called(*a, **kw):
raise AssertionError("siteverify should not run when no secret is configured")
monkeypatch.setattr(turnstile_mod.httpx, "post", must_not_be_called)
app, _fake = app_with_fake_gitea
with TestClient(app) as client:
_reset_outbound()
r = client.post(
"/auth/otc/request",
json={"email": "alice@example.com"},
)
assert r.status_code == 200, r.text
# The OTC path ran end-to-end — one envelope to the address.
assert len(_outbound_otc_envelopes("alice@example.com")) == 1
# ---------------------------------------------------------------------------
# Missing-secret-hard: no secret, TURNSTILE_REQUIRED=true → 500 "misconfigured"
# ---------------------------------------------------------------------------
def test_otc_request_refuses_when_required_but_secret_unset(app_with_fake_gitea, monkeypatch):
"""Once the operator has flipped `TURNSTILE_REQUIRED=true` to lock
down production, a missing secret stops being a soft-fail and
becomes a fail-closed 500. This is the regression-detection shape
the §20.4 upgrade-steps MAY block calls out flip the flag once
the secret is wired so a future config drift fails loudly instead
of silently disabling abuse defense.
"""
from fastapi.testclient import TestClient
monkeypatch.delenv("CLOUDFLARE_TURNSTILE_SECRET", raising=False)
monkeypatch.setenv("TURNSTILE_REQUIRED", "true")
app, _fake = app_with_fake_gitea
with TestClient(app) as client:
_reset_outbound()
r = client.post(
"/auth/otc/request",
json={"email": "alice@example.com", "turnstile_token": "doesnt-matter"},
)
assert r.status_code == 500, r.text
assert _outbound_otc_envelopes("alice@example.com") == []
+13
View File
@@ -49,3 +49,16 @@ VITE_PRIVACY_POLICY_URL=
# Examples:
# VITE_COOKIES_POLICY_URL=https://wiggleverse.org/cookies
VITE_COOKIES_POLICY_URL=
# v0.12.0 / roadmap item #10: CloudFlare Turnstile site key (public).
# Provision a Turnstile site at dash.cloudflare.com → Turnstile → Add
# site. The site key (this var) is embedded into the frontend bundle at
# build time and rendered by the Turnstile widget on the /login email-
# entry step. The secret key (private) lives in the backend env as
# CLOUDFLARE_TURNSTILE_SECRET — see backend/.env.example. Leave unset
# in dev to skip the widget; the backend's TURNSTILE_REQUIRED policy
# decides what happens to a tokenless request.
#
# Examples:
# VITE_TURNSTILE_SITE_KEY=0x4AAAAAAA...
VITE_TURNSTILE_SITE_KEY=
+2 -2
View File
@@ -1,12 +1,12 @@
{
"name": "rfc-app-frontend",
"version": "0.10.0",
"version": "0.12.0",
"lockfileVersion": 3,
"requires": true,
"packages": {
"": {
"name": "rfc-app-frontend",
"version": "0.10.0",
"version": "0.12.0",
"dependencies": {
"@codemirror/commands": "^6.10.3",
"@codemirror/lang-markdown": "^6.5.0",
+1 -1
View File
@@ -1,7 +1,7 @@
{
"name": "rfc-app-frontend",
"private": true,
"version": "0.10.0",
"version": "0.12.0",
"type": "module",
"scripts": {
"dev": "vite",
+143
View File
@@ -410,6 +410,26 @@
cursor: pointer; padding: 0;
}
.otc-login .btn-link-quiet:hover { color: #1a1a1a; text-decoration: underline; }
/* v0.8.0 — labels + textarea for the first-OTC profile capture step. */
.otc-field-label {
font-size: 12px; color: #666;
margin: 8px 0 -4px;
font-weight: 600;
}
.otc-login textarea {
width: 100%;
padding: 10px 12px;
font-size: 15px;
border: 1px solid #ddd;
border-radius: 6px;
box-sizing: border-box;
font-family: inherit;
resize: vertical;
}
.otc-login textarea:focus {
outline: none;
border-color: #1a1a1a;
}
.otc-shortcut-hint {
color: #888; font-size: 12px; margin: 4px 0 0;
}
@@ -435,6 +455,65 @@
.otc-fallback a:hover { color: #1a1a1a; text-decoration: underline; }
.otc-fallback-sep { color: #ccc; }
/* v0.11.0 "trust this device for 30 days" checkbox on the verify
step. Sits above the action row, padded so it doesn't crowd the
passcode/code input. */
.otc-trust-device {
display: flex; align-items: center; gap: 8px;
font-size: 13px; color: #444;
margin: 8px 0 4px;
cursor: pointer;
user-select: none;
}
.otc-trust-device input[type="checkbox"] {
width: auto; margin: 0; cursor: pointer;
}
/* v0.11.0 — /settings/devices revoke-device UI. */
.device-list {
list-style: none; padding: 0; margin: 12px 0 0;
}
.device-list-item {
display: flex; align-items: center; justify-content: space-between;
gap: 12px;
border: 1px solid #eee; border-radius: 6px;
padding: 10px 12px; margin: 0 0 8px;
background: #fafafa;
}
.device-list-item .device-meta {
flex: 1; min-width: 0;
}
.device-list-item .device-ua {
font-size: 13px; color: #1a1a1a;
white-space: nowrap; overflow: hidden; text-overflow: ellipsis;
}
.device-list-item .device-stamps {
font-size: 12px; color: #777;
margin-top: 2px;
}
.device-list-item button {
font-size: 12px; padding: 4px 10px;
border: 1px solid #ccc; border-radius: 4px;
background: white; cursor: pointer;
}
.device-list-item button:hover:not(:disabled) {
background: #f5f5f5;
}
.device-revoke-all {
margin-top: 8px;
font-size: 13px; padding: 6px 12px;
border: 1px solid #cb6a6a; border-radius: 4px;
background: white; color: #cb6a6a; cursor: pointer;
}
.device-revoke-all:hover:not(:disabled) {
background: #fff5f5;
}
.device-empty {
font-size: 13px; color: #777;
background: #fafafa; border: 1px solid #eee; border-radius: 6px;
padding: 12px;
}
/* --- Beta-pending page (post-OAuth-rejection) --- */
.beta-pending {
@@ -466,6 +545,22 @@
.btn-link-quiet { color: #666; text-decoration: none; font-size: 13px; }
.btn-link-quiet:hover { color: #1a1a1a; text-decoration: underline; }
/* v0.8.0 thin "your beta access is in review" banner. Shown on every
page (other than /beta-pending itself, which carries the larger
form of the message). Sits just under the app header so it doesn't
compete with the catalog rail. */
.pending-access-banner {
background: #fff8e0;
border-bottom: 1px solid #e6dca0;
color: #4a3f00;
font-size: 13px;
padding: 8px 16px;
text-align: center;
}
.pending-access-banner a {
color: #4a3f00; text-decoration: underline;
}
/* ── §8 RFC view: three-column shape ─────────────────────────────────── */
.main-pane {
@@ -1853,6 +1948,54 @@
display: inline-flex; align-items: center; gap: 6px;
font-size: 13px; cursor: pointer;
}
/* v0.9.0 — admin user-management surface (roadmap item #7). */
.admin-filter-chips {
display: flex; gap: 6px; margin-bottom: 16px; flex-wrap: wrap;
}
.admin-chip {
display: inline-flex; align-items: center; gap: 6px;
background: #fff; border: 1px solid #d1d5db; border-radius: 999px;
padding: 4px 12px; font-size: 12px; color: #374151; cursor: pointer;
}
.admin-chip:hover { background: #f9fafb; }
.admin-chip.active {
background: #111; color: #fff; border-color: #111;
}
.admin-chip-count {
font-size: 11px; opacity: 0.7;
}
.admin-users-table td { vertical-align: top; padding-top: 10px; padding-bottom: 10px; }
.permission-cell { display: flex; flex-direction: column; gap: 4px; }
.permission-actions { display: flex; gap: 6px; }
.permission-badge {
display: inline-block;
font-size: 11px; font-weight: 600;
padding: 2px 8px; border-radius: 999px;
text-transform: uppercase; letter-spacing: 0.04em;
width: max-content;
}
.permission-badge-pending {
background: #fef3c7; color: #92400e;
}
.permission-badge-granted {
background: #dcfce7; color: #166534;
}
.permission-badge-revoked {
background: #fee2e2; color: #991b1b;
}
.permission-decided { font-size: 11px; }
.user-row-reason td {
background: #fffbeb; border-top: none !important;
padding: 0 16px 12px !important;
}
.user-reason-block {
border-left: 3px solid #f59e0b;
padding: 8px 12px; font-size: 13px;
background: #fffbeb;
}
.user-reason-block strong { display: block; margin-bottom: 4px; color: #92400e; }
.user-reason-block p { margin: 0; white-space: pre-wrap; color: #374151; }
.grad-queue { list-style: none; padding: 0; margin: 8px 0 24px; }
.grad-queue li { padding: 8px 0; border-bottom: 1px solid #f3f4f6; }
.grad-queue-link { color: #111; text-decoration: none; font-size: 14px; }
+41 -4
View File
@@ -11,6 +11,7 @@ import Landing from './components/Landing.jsx'
import Login from './components/Login.jsx'
import BetaPending from './components/BetaPending.jsx'
import Philosophy from './components/Philosophy.jsx'
import Docs from './components/Docs.jsx'
import NotificationSettings from './components/NotificationSettings.jsx'
import Admin from './components/Admin.jsx'
import ToastHost, { showToast } from './components/ToastHost.jsx'
@@ -87,11 +88,15 @@ export default function App() {
// The deployment is in private beta: anonymous visitors get the full
// app in read-only mode (viewer = null is passed through to every
// component), and write affordances are hidden at the component
// level. /beta-pending is the post-OAuth-rejection page reachable by
// anyone. The original §14.1 Landing surface is retained for the
// `/welcome` URL only, in case a deployment wants to link to it.
// level. v0.8.0 (§6.1 / item #6): authenticated users with
// `permission_state='pending'` also pass through as `viewer` with
// their state attached every write-gated affordance reads the
// state and treats pending the same as anonymous, while reads
// remain open. The /beta-pending page is the home root for a
// pending user.
const viewer = me?.authenticated ? me.user : null
const isAdmin = viewer && (viewer.role === 'owner' || viewer.role === 'admin')
const isPending = viewer && viewer.permission_state === 'pending'
return (
<div className="app">
@@ -107,6 +112,9 @@ export default function App() {
<Link to="/philosophy" className="header-about" title="Why this exists (§14)">
About
</Link>
<Link to="/docs" className="header-about" title="User guide">
Docs
</Link>
{viewer && (
<Link to="/settings/notifications" className="header-settings" title="Notification settings (§15)">
Settings
@@ -142,12 +150,14 @@ export default function App() {
)}
</div>
</header>
{isPending && <PendingAccessBanner />}
<div className="app-body">
<Routes>
<Route path="/welcome" element={<Landing />} />
<Route path="/login" element={<Login />} />
<Route path="/beta-pending" element={<BetaPending />} />
<Route path="/beta-pending" element={<BetaPending viewer={viewer} />} />
<Route path="/philosophy" element={<PhilosophyWithSidebar viewer={viewer} />} />
<Route path="/docs" element={<DocsWithSidebar viewer={viewer} />} />
{/* §14.5 / §14.6: cookie-consent companions to /philosophy.
Available to anonymous and authenticated viewers alike. */}
<Route path="/privacy" element={<PolicyShell><Privacy /></PolicyShell>} />
@@ -216,6 +226,14 @@ function PhilosophyWithSidebar({ viewer }) {
)
}
function DocsWithSidebar({ viewer }) {
return (
<main className="chrome-pane">
<Docs authenticated={!!viewer} />
</main>
)
}
function NotificationSettingsWithSidebar({ viewer }) {
return (
<main className="chrome-pane">
@@ -232,7 +250,26 @@ function AdminWithSidebar({ viewer }) {
)
}
function PendingAccessBanner() {
// v0.8.0 thin banner shown on every page (other than /beta-pending
// itself, which carries the same message in larger form) when the
// signed-in user's `permission_state='pending'`. Sign-out works
// normally via the header affordance.
return (
<div className="pending-access-banner">
Your beta access request is in review.{' '}
<Link to="/beta-pending">Learn more </Link>
</div>
)
}
function Welcome({ viewer }) {
// v0.8.0 a pending user landing on "/" gets the same page they'd
// see at /beta-pending, inline. This is the post-OTC home root for
// a user awaiting admin grant.
if (viewer && viewer.permission_state === 'pending') {
return <BetaPending viewer={viewer} />
}
if (!viewer) {
return (
<div className="welcome">
+81 -6
View File
@@ -31,20 +31,49 @@ export async function getMe() {
// migration — the new UI just no longer points at it primarily. These
// two helpers drive the Login.jsx surface.
export async function requestOtc(email) {
export async function requestOtc(email, { turnstileToken } = {}) {
// v0.12.0 / roadmap item #10: when the Turnstile widget has produced
// a token, send it alongside the email so the backend can siteverify
// before the OTC dispatch. The backend treats a missing token as
// either soft-fail (no secret wired AND TURNSTILE_REQUIRED=false)
// or hard-fail (verification required) — the frontend stays
// uninvolved in the policy.
const body = { email }
if (turnstileToken) body.turnstile_token = turnstileToken
const res = await fetch('/auth/otc/request', {
method: 'POST',
headers: { 'Content-Type': 'application/json' },
body: JSON.stringify({ email }),
body: JSON.stringify(body),
})
return jsonOrThrow(res)
}
export async function verifyOtc(email, code) {
export async function verifyOtc(email, code, { trustDevice = false } = {}) {
// v0.11.0 — `trustDevice` is the "trust this device for 30 days"
// checkbox on the Login.jsx OTC step. When true, the server mints
// a fresh device-trust row and sets the long-lived cookie; on
// subsequent visits, the cookie skips the OTC roundtrip via
// `startDeviceTrust()`.
const res = await fetch('/auth/otc/verify', {
method: 'POST',
headers: { 'Content-Type': 'application/json' },
body: JSON.stringify({ email, code }),
body: JSON.stringify({ email, code, trust_device: !!trustDevice }),
})
return jsonOrThrow(res)
}
// ── v0.8.0: open beta-access request flow (§6.1 / §14.1) ─────────────────
//
// On the first OTC sign-in, the user lands in `permission_state='pending'`
// and `/api/auth/me` reports `needs_profile=true`. The Login.jsx surface
// then prompts for first/last/why and POSTs them here. After this lands,
// the user sees the /beta-pending page until an admin grants access.
export async function submitBetaRequest({ first_name, last_name, beta_request_reason }) {
const res = await fetch('/api/auth/me/beta-request', {
method: 'POST',
headers: { 'Content-Type': 'application/json' },
body: JSON.stringify({ first_name, last_name, beta_request_reason }),
})
return jsonOrThrow(res)
}
@@ -66,15 +95,44 @@ export async function checkPasscode(email) {
return jsonOrThrow(res)
}
export async function verifyPasscode(email, passcode) {
export async function verifyPasscode(email, passcode, { trustDevice = false } = {}) {
// v0.11.0 — same trust-device opt-in as `verifyOtc`.
const res = await fetch('/auth/passcode/verify', {
method: 'POST',
headers: { 'Content-Type': 'application/json' },
body: JSON.stringify({ email, passcode }),
body: JSON.stringify({ email, passcode, trust_device: !!trustDevice }),
})
return jsonOrThrow(res)
}
// ── v0.11.0: trust device for 30 days (§6.2, roadmap item #9) ─────────────
//
// On a returning visit with a valid device-trust cookie, `startDeviceTrust`
// re-establishes the session without an OTC / passcode roundtrip. The
// cookie is HttpOnly so the client cannot read it; the call is a pure POST
// that the browser attaches the cookie to automatically.
//
// `listMyDevices`, `revokeMyDevice`, and `revokeAllMyDevices` drive the
// /settings/devices revoke-device UI. The signed-in user is the implicit
// subject; the cookie carries the session.
export async function startDeviceTrust() {
const res = await fetch('/auth/device-trust/start', { method: 'POST' })
return jsonOrThrow(res)
}
export async function listMyDevices() {
return jsonOrThrow(await fetch('/api/auth/me/devices'))
}
export async function revokeMyDevice(deviceId) {
return jsonOrThrow(await fetch(`/api/auth/me/devices/${deviceId}`, { method: 'DELETE' }))
}
export async function revokeAllMyDevices() {
return jsonOrThrow(await fetch('/api/auth/me/devices', { method: 'DELETE' }))
}
export async function setPasscode(passcode) {
// Requires an active session — the server returns 401 if not signed
// in. The signed-in user is the implicit subject; the body carries
@@ -616,6 +674,10 @@ export async function getPhilosophy() {
return jsonOrThrow(await fetch('/api/philosophy'))
}
export async function getDocs() {
return jsonOrThrow(await fetch('/api/docs'))
}
// ---------------------------------------------------------------------------
// Slice 7: admin neighborhood (§17 admin/* + user search for the §15.8 mute
// typeahead).
@@ -641,6 +703,19 @@ export async function setUserMute(userId, muted) {
}))
}
// v0.9.0 — roadmap item #7. Flip a user's permission_state between
// 'pending', 'granted', and 'revoked'. The Users tab on the admin
// page wires Grant / Revoke buttons against this endpoint; the
// returned `changed` flag is false when the requested state already
// matched the row.
export async function setUserPermission(userId, state) {
return jsonOrThrow(await fetch(`/api/admin/users/${userId}/permission`, {
method: 'POST',
headers: { 'Content-Type': 'application/json' },
body: JSON.stringify({ state }),
}))
}
export async function listAuditLog({ actionKind, actorUserId, rfcSlug, beforeId, limit } = {}) {
const params = new URLSearchParams()
if (actionKind) params.set('action_kind', actionKind)
+192 -52
View File
@@ -16,6 +16,7 @@ import {
listAdminUsers,
setUserRole,
setUserMute,
setUserPermission,
listAuditLog,
listPermissionEvents,
listGraduationQueue,
@@ -68,12 +69,26 @@ export default function Admin({ viewer }) {
)
}
// Users + role + write-mute (§6.1 / §6.2)
// Users + role + write-mute + permission grant/revoke (§6.1 / §6.2)
//
// v0.9.0 (roadmap item #7) lands the user-management surface. The table
// shows every user with their permission_state, sign-up reason (when
// pending), role, write-mute, and Grant / Revoke controls. State filter
// chips above the table narrow to one bucket the "Pending" chip is the
// admin's daily inbox shape.
const STATE_CHIPS = [
{ value: 'all', label: 'All' },
{ value: 'pending', label: 'Pending' },
{ value: 'granted', label: 'Granted' },
{ value: 'revoked', label: 'Revoked' },
]
function UsersTab() {
const [users, setUsers] = useState(null)
const [busy, setBusy] = useState({})
const [error, setError] = useState(null)
const [stateFilter, setStateFilter] = useState('all')
async function refresh() {
setError(null)
@@ -113,68 +128,193 @@ function UsersTab() {
}
}
async function flipPermission(userId, state) {
setBusy(b => ({ ...b, [userId]: true }))
setError(null)
try {
await setUserPermission(userId, state)
// Refresh the full row so permission_decided_{at,by_*} update too.
await refresh()
} catch (e) {
setError(e.message)
} finally {
setBusy(b => ({ ...b, [userId]: false }))
}
}
const counts = useMemo(() => {
const c = { all: 0, pending: 0, granted: 0, revoked: 0 }
if (users) {
c.all = users.length
for (const u of users) {
const s = u.permission_state || 'granted'
if (s in c) c[s] += 1
}
}
return c
}, [users])
if (users == null) return <p className="muted">Loading users</p>
const filtered = stateFilter === 'all'
? users
: users.filter(u => (u.permission_state || 'granted') === stateFilter)
return (
<div className="admin-tab">
<header className="admin-tab-header">
<h2>Users</h2>
<p className="muted">
Role changes write to <code>permission_events</code>. The §6.2
write-mute applies to contributors only promote to admin to
remove a user's ability to write without silencing them.
The pending bucket is the beta-access review queue (§6.1 /
v0.8.0). Grant or revoke writes to <code>permission_events</code>
and stamps <code>permission_decided_by</code> +{' '}
<code>permission_decided_at</code>. Role and write-mute controls
retain their v0.7.0 semantics promote to admin to remove a
user's ability to write without silencing them.
</p>
</header>
{error && <p className="settings-note warning">{error}</p>}
<table className="admin-table">
<thead>
<tr>
<th>User</th>
<th>Role</th>
<th>Write-muted</th>
<th>Last seen</th>
</tr>
</thead>
<tbody>
{users.map(u => (
<tr key={u.id}>
<td>
<div className="user-cell">
<span className="user-handle">@{u.gitea_login}</span>
<span className="muted">{u.display_name}</span>
</div>
</td>
<td>
<select
value={u.role}
onChange={e => changeRole(u.id, e.target.value)}
disabled={!!busy[u.id]}
>
<option value="contributor">Contributor</option>
<option value="admin">Admin</option>
<option value="owner">Owner</option>
</select>
</td>
<td>
{u.role === 'contributor' ? (
<label className="mute-toggle">
<input
type="checkbox"
checked={!!u.muted}
onChange={e => toggleMute(u.id, e.target.checked)}
disabled={!!busy[u.id]}
/>
{u.muted ? 'Muted' : 'Active'}
</label>
) : (
<span className="muted">N/A</span>
)}
</td>
<td className="muted">{u.last_seen_at}</td>
<div className="admin-filter-chips">
{STATE_CHIPS.map(chip => (
<button
key={chip.value}
type="button"
className={`admin-chip${stateFilter === chip.value ? ' active' : ''}`}
onClick={() => setStateFilter(chip.value)}
>
{chip.label} <span className="admin-chip-count">{counts[chip.value] ?? 0}</span>
</button>
))}
</div>
{filtered.length === 0 ? (
<p className="muted">No users in this bucket.</p>
) : (
<table className="admin-table admin-users-table">
<thead>
<tr>
<th>User</th>
<th>State</th>
<th>Role</th>
<th>Write-muted</th>
<th>Signed up</th>
<th>Last seen</th>
</tr>
))}
</tbody>
</table>
</thead>
<tbody>
{filtered.map(u => (
<UserRow
key={u.id}
user={u}
busy={!!busy[u.id]}
onChangeRole={role => changeRole(u.id, role)}
onToggleMute={muted => toggleMute(u.id, muted)}
onFlipPermission={state => flipPermission(u.id, state)}
/>
))}
</tbody>
</table>
)}
</div>
)
}
function UserRow({ user: u, busy, onChangeRole, onToggleMute, onFlipPermission }) {
const state = u.permission_state || 'granted'
const fullName = [u.first_name, u.last_name].filter(Boolean).join(' ').trim()
const handle = u.gitea_login ? `@${u.gitea_login}` : (u.email || u.display_name)
return (
<>
<tr>
<td>
<div className="user-cell">
<span className="user-handle">{handle}</span>
<span className="muted">
{fullName || u.display_name}
{u.email ? ` · ${u.email}` : ''}
</span>
</div>
</td>
<td>
<PermissionCell user={u} busy={busy} onFlipPermission={onFlipPermission} />
</td>
<td>
<select
value={u.role}
onChange={e => onChangeRole(e.target.value)}
disabled={busy}
>
<option value="contributor">Contributor</option>
<option value="admin">Admin</option>
<option value="owner">Owner</option>
</select>
</td>
<td>
{u.role === 'contributor' ? (
<label className="mute-toggle">
<input
type="checkbox"
checked={!!u.muted}
onChange={e => onToggleMute(e.target.checked)}
disabled={busy}
/>
{u.muted ? 'Muted' : 'Active'}
</label>
) : (
<span className="muted">N/A</span>
)}
</td>
<td className="muted">{u.created_at || '—'}</td>
<td className="muted">{u.last_seen_at || '—'}</td>
</tr>
{state === 'pending' && u.beta_request_reason ? (
<tr className="user-row-reason">
<td colSpan={6}>
<div className="user-reason-block">
<strong>Why they want access:</strong>
<p>{u.beta_request_reason}</p>
</div>
</td>
</tr>
) : null}
</>
)
}
function PermissionCell({ user: u, busy, onFlipPermission }) {
const state = u.permission_state || 'granted'
const decidedSuffix = u.permission_decided_at
? ` · by ${u.permission_decided_by_login ? '@' + u.permission_decided_by_login : '—'} at ${u.permission_decided_at}`
: ''
return (
<div className="permission-cell">
<span className={`permission-badge permission-badge-${state}`}>{state}</span>
<div className="permission-actions">
{state !== 'granted' && (
<button
type="button"
className="btn-link-quiet"
disabled={busy}
onClick={() => onFlipPermission('granted')}
>Grant</button>
)}
{state === 'granted' && (
<button
type="button"
className="btn-link-quiet"
disabled={busy}
onClick={() => {
if (confirm(`Revoke access for ${u.display_name || u.email}?`)) {
onFlipPermission('revoked')
}
}}
>Revoke</button>
)}
</div>
{decidedSuffix && (
<div className="permission-decided muted">{decidedSuffix.replace(/^ · /, '')}</div>
)}
</div>
)
}
+43 -19
View File
@@ -1,38 +1,62 @@
// BetaPending.jsx the post-OAuth-rejection page.
// BetaPending.jsx the "your request is in review" page (§6.1 / §14.1).
//
// When a deployment is in private-beta mode (i.e. its `allowed_emails`
// table has any rows), the OAuth callback redirects unrecognised users
// here instead of provisioning them. The framework cannot know the
// deployment operator's preferred contact channel so the deployment
// supplies one via VITE_BETA_CONTACT (an email, URL, or short
// instruction). If unset, we render a generic ask-the-operator line.
// v0.3.0 introduced this surface as the post-OAuth-rejection page (a
// user whose email wasn't on the `allowed_emails` table bounced here).
// v0.8.0 (roadmap item #6) repurposes it as the post-OTC pending-grant
// page: any authenticated user whose `permission_state='pending'` lands
// here on root visits, after a fresh-OTC profile capture, or via the
// header "Your beta access is in review" affordance.
//
// The deployment supplies a contact channel via VITE_BETA_CONTACT (an
// email, URL, or short instruction). If unset, we render a generic
// ask-the-operator line.
import { Link } from 'react-router-dom'
export default function BetaPending() {
export default function BetaPending({ viewer }) {
const contact = import.meta.env.VITE_BETA_CONTACT || ''
const isPending = viewer?.permission_state === 'pending'
return (
<div className="beta-pending">
<div className="beta-pending-inner">
<h1>{import.meta.env.VITE_APP_NAME} is in private Beta.</h1>
<p>
Discussion and contribution are gated to invited emails for now.
Reading is open every super-draft, every active RFC, and every
public conversation is visible without signing in.
</p>
<h1>
{isPending
? 'Your request is in review.'
: `${import.meta.env.VITE_APP_NAME} is in private Beta.`}
</h1>
{isPending ? (
<>
<p>
Thanks for telling us a bit about yourself. The deployment's
admins are notified by email as soon as a request lands;
we don't commit to a fixed SLA turnaround depends on
operator availability and the deployment operator is
the right person to ask if a wait runs long.
</p>
<p>
While you wait, the catalog on the left lists every super-draft
and active RFC in the framework reading is open. Discussion
and contribution unlock once your access is granted.
</p>
</>
) : (
<p>
Discussion and contribution are gated to invited contributors for
now. Reading is open every super-draft, every active RFC, and
every public conversation is visible without signing in.
</p>
)}
{contact ? (
<p className="beta-pending-contact">
To request access, contact <strong>{contact}</strong> with the
email address you'd like to sign in with.
Questions? Contact <strong>{contact}</strong>.
</p>
) : (
<p className="beta-pending-contact">
To request access, contact the deployment operator with the email
address you'd like to sign in with.
Questions? Contact the deployment operator.
</p>
)}
<div className="beta-pending-actions">
<Link className="btn-primary" to="/">Browse as a guest</Link>
<Link className="btn-primary" to="/">Browse the catalog</Link>
<Link className="btn-link-quiet" to="/philosophy">Read the philosophy </Link>
</div>
</div>
+51
View File
@@ -0,0 +1,51 @@
// `/docs` the user-facing guide.
//
// Sibling of Philosophy.jsx: same chrome, same data path, different
// source file. Renders DOCS.md verbatim with light chrome around it.
// Reachable anonymously, same as `/philosophy`, so a visitor can read
// the guide before deciding to sign in.
import { useEffect, useState } from 'react'
import { Link, useNavigate } from 'react-router-dom'
import MarkdownPreview from './MarkdownPreview.jsx'
import { getDocs } from '../api.js'
export default function Docs({ authenticated }) {
const [body, setBody] = useState('')
const [error, setError] = useState(null)
const [loading, setLoading] = useState(true)
const navigate = useNavigate()
useEffect(() => {
let active = true
getDocs()
.then(r => { if (active) setBody(r.body || '') })
.catch(e => { if (active) setError(e.message || String(e)) })
.finally(() => { if (active) setLoading(false) })
return () => { active = false }
}, [])
return (
<div className="philosophy-page">
<header className="philosophy-header">
<button
className="philosophy-back"
onClick={() => (history.length > 1 ? navigate(-1) : navigate('/'))}
>
Back
</button>
<span className="philosophy-title">User guide</span>
{!authenticated && (
<Link className="philosophy-signin" to="/">Home</Link>
)}
</header>
<article className="philosophy-body">
{loading && <p className="muted">Loading</p>}
{error && <p className="error">Could not load the guide: {error}</p>}
{!loading && !error && (
<MarkdownPreview content={body} />
)}
</article>
</div>
)
}
+351 -56
View File
@@ -1,71 +1,163 @@
// Login.jsx v0.7.0's email + OTC sign-in surface (§6.2), extended
// in v0.10.0 with passcode sign-in (roadmap item #8).
// Login.jsx the composed sign-in surface (§6.2) after the v0.12.0
// (CloudFlare Turnstile gate on OTC dispatch, roadmap item #10) /
// v0.10.0 (passcodes, roadmap item #8) rebase onto v0.8.0
// (beta-access-request capture, §6.1 / §14.1, roadmap item #6). v0.7.0
// (roadmap item #5) established the email + OTC scaffolding the later
// releases extended.
//
// Three-to-five-step flow:
// 1. Enter email GET /auth/passcode/check.
// * If `has_passcode`: advance to step 'passcode'.
// * Otherwise: POST /auth/otc/request, advance to step 'code'.
// 2a. Step 'passcode': enter passcode POST /auth/passcode/verify.
// * On 200: redirect to /.
// * On 423: passcode is locked (5 consecutive failures); the
// UI auto-falls back to OTC by requesting a fresh code.
// * On 400: wrong passcode; the user can retry or click
// "Use a code instead" to fall back to OTC manually.
// 2b. Step 'code' (the v0.7.0 path): enter the six-digit code
// POST /auth/otc/verify on 200, the server has signed in the
// user. If the user has no passcode set, we show step
// 'offer-passcode' inviting them to set one for faster sign-in
// next time. Dismiss skips to /; "Set passcode" advances to
// step 'set-passcode'.
// 3. Step 'set-passcode': enter a passcode POST /auth/passcode/set
// redirect to /. The user can also "Skip for now".
// v0.12.0: the email-entry step renders a Turnstile widget. The token
// it produces is sent to `/auth/otc/request` alongside the email. The
// passcode step also renders a widget for the "Use a code instead"
// fallback dispatch (same backend endpoint, same gate). When the
// `VITE_TURNSTILE_SITE_KEY` build var is unset, the widget renders
// nothing the form still submits and the backend's TURNSTILE_REQUIRED
// policy decides admission.
//
// Server-side, /auth/otc/request always returns 202 for an unrecognized
// email (so the allowlist gate doesn't leak), so this surface never
// distinguishes "we couldn't reach you" from "we don't know you". The
// check endpoint also returns `has_passcode: false` for an unknown
// email so an unknown email always lands in the OTC path, no
// account-enumeration signal.
// Four-to-six-step flow (most users see three; the longest path is
// pending-user with no passcode, who never sees the passcode steps):
//
// The legacy Gitea OAuth callback remains at /auth/login /auth/callback
// during the v0.7.0 migration; we surface a "Sign in with Gitea" link
// as a fallback in the footer so users with active OAuth sessions or
// older invite emails still have a path.
// 1. 'email' Enter email GET /auth/passcode/check.
// * has_passcode=true step 'passcode'.
// * has_passcode=false POST /auth/otc/request,
// step 'code'.
// 429 on either dispatch surfaces a "wait a
// moment" hint and keeps the user on step 1.
//
// 2a. 'passcode' Enter passcode POST /auth/passcode/verify.
// * 200 redirect to "/".
// * 423 (lockout, 5 consecutive failures)
// auto-fall back to OTC by requesting a fresh
// code and advancing to step 'code'.
// * 400 wrong passcode; user can retry or
// click "Use a code instead" to fall back
// manually.
//
// 2b. 'code' Enter the six-digit code POST /auth/otc/verify.
// On 200, fetch /api/auth/me and branch:
// * needs_profile === true 'capture-profile'
// * has_passcode === false 'offer-passcode'
// * otherwise redirect to "/".
// needs_profile WINS over has_passcode a
// pending user goes through the §6.1 capture
// flow first; setting a passcode while waiting
// for admin grant gains them nothing.
// Cmd/Ctrl+Enter on the code field is the
// keyboard shortcut.
//
// 3. 'capture-profile' (v0.8.0, §6.1) First name, last name, and "why
// I should be included in the beta" POST
// /api/auth/me/beta-request redirect to
// /beta-pending. The user's row stays
// permission_state='pending' until an admin
// grants access; they can set a passcode later
// from settings, or on a future sign-in once
// granted.
//
// 4a. 'offer-passcode' (v0.10.0) "Set a passcode for faster sign-in
// next time?" Yes 'set-passcode'. Skip "/".
//
// 4b. 'set-passcode' Pick a passcode (420 chars) POST
// /auth/passcode/set redirect to "/". A
// "Skip for now" link also redirects to "/".
//
// Server-side, /auth/otc/request returns 202 uniformly and
// /auth/passcode/check returns has_passcode=false for an unknown
// email, so this surface never distinguishes "we couldn't reach you"
// from "we don't know you" an unknown email always lands in the
// OTC path with no account-enumeration signal.
//
// The legacy Gitea OAuth callback remains at /auth/login
// /auth/callback during the v0.7.0 migration; we surface a "Sign in
// with Gitea" link as a fallback in the footer so users with active
// OAuth sessions or older invite emails still have a path. We hide
// the fallback on 'capture-profile' so a half-captured pending user
// doesn't bail out into the OAuth path mid-form.
import { useEffect, useRef, useState } from 'react'
import { useNavigate, Link } from 'react-router-dom'
import {
requestOtc,
verifyOtc,
submitBetaRequest,
checkPasscode,
verifyPasscode,
setPasscode as apiSetPasscode,
startDeviceTrust,
} from '../api'
import TurnstileWidget, { turnstileEnabled } from './TurnstileWidget'
export default function Login() {
// Steps: 'email' 'passcode' or 'code' (after OTC verify) optional
// 'offer-passcode' optional 'set-passcode'. The latter two only
// appear on the OTC path for accounts that don't yet have a passcode.
// Steps: 'email' 'passcode' or 'code' (on the OTC path, after
// verify) one of: 'capture-profile' (pending user), 'offer-passcode'
// (no passcode yet), or straight to "/". 'set-passcode' is reached
// from 'offer-passcode'.
const [step, setStep] = useState('email')
const [email, setEmail] = useState('')
const [code, setCode] = useState('')
const [passcode, setPasscode] = useState('')
const [newPasscode, setNewPasscode] = useState('')
// v0.11.0 "trust this device for 30 days" checkbox, shared by the
// OTC and passcode verify steps. The flag rides on the verify POST;
// a checked box mints a device-trust row server-side and sets the
// long-lived `rfc_device_trust` cookie. Defaults off so the user
// makes an explicit choice auth credentials shouldn't persist by
// default.
const [trustDevice, setTrustDevice] = useState(false)
// v0.8.0 capture-profile fields.
const [firstName, setFirstName] = useState('')
const [lastName, setLastName] = useState('')
const [reason, setReason] = useState('')
const [status, setStatus] = useState('')
const [busy, setBusy] = useState(false)
// v0.12.0: Turnstile token captured by the widget. `null` means no
// challenge solved yet (or the site key is unset, in which case the
// widget surfaces null on mount). The token is single-use; we clear
// it back to null right after we send it so a second request on the
// same form remount re-challenges. `turnstileReady` is true once the
// widget has produced a token OR the widget is not configured at
// build time (no site key) the submit button reads from it so the
// form locks up when the operator has wired Turnstile but the user
// hasn't solved the challenge yet.
const [turnstileToken, setTurnstileToken] = useState(null)
const turnstileOn = turnstileEnabled()
const turnstileReady = !turnstileOn || !!turnstileToken
const emailRef = useRef(null)
const codeRef = useRef(null)
const passcodeRef = useRef(null)
const newPasscodeRef = useRef(null)
const firstNameRef = useRef(null)
const navigate = useNavigate()
useEffect(() => {
if (step === 'email') emailRef.current?.focus()
else if (step === 'code') codeRef.current?.focus()
else if (step === 'passcode') passcodeRef.current?.focus()
else if (step === 'capture-profile') firstNameRef.current?.focus()
else if (step === 'set-passcode') newPasscodeRef.current?.focus()
}, [step])
// v0.11.0 on mount, try the device-trust cookie path. If the
// browser still carries a valid `rfc_device_trust` cookie from a
// prior "trust this device" gesture, the server re-establishes the
// session without any user input and we redirect home. The cookie
// is HttpOnly so we can't peek at it; we just call the endpoint and
// see whether it returns 200. 401 (no cookie / invalid / revoked)
// is the structural-silent case the user proceeds to the email
// step normally. We do not surface any UI about the attempt; a
// failure should be invisible.
useEffect(() => {
let cancelled = false
;(async () => {
try {
await startDeviceTrust()
if (!cancelled) window.location.assign('/')
} catch (_) {
// No trusted device fall through to the email step.
}
})()
return () => { cancelled = true }
}, [])
async function submitEmail(e) {
e.preventDefault()
if (!email.trim() || !email.includes('@')) {
@@ -80,13 +172,22 @@ export default function Login() {
setStep('passcode')
setStatus('')
} else {
await requestOtc(email.trim())
await requestOtc(email.trim(), { turnstileToken })
// v0.12.0: the token is single-use; drop it so a re-request
// from the code step (via "Use a different email" back to
// email) starts with a fresh challenge.
setTurnstileToken(null)
setStep('code')
setStatus('Check your inbox — a six-digit code is on the way.')
}
} catch (err) {
// Any failure consumes the token from CloudFlare's side; clear
// so the widget re-renders a fresh challenge on retry.
setTurnstileToken(null)
if (err.status === 429) {
setStatus('Slow down — wait a minute before requesting another code.')
} else if (err.status === 400) {
setStatus("Couldn't verify you're human. Please retry the challenge.")
} else {
setStatus(err.message || 'Could not start sign-in. Try again.')
}
@@ -104,7 +205,11 @@ export default function Login() {
setBusy(true)
setStatus('')
try {
await verifyPasscode(email.trim(), passcode.trim())
await verifyPasscode(email.trim(), passcode.trim(), { trustDevice })
// Reload so App.jsx's getMe() picks up the fresh session. A
// returning passcode user is by definition already past the
// §6.1 capture step (they couldn't have set a passcode while
// pending), so we go straight to "/".
window.location.assign('/')
} catch (err) {
if (err.status === 423) {
@@ -113,17 +218,29 @@ export default function Login() {
// fresh code in the user's inbox immediately.
setPasscode('')
try {
await requestOtc(email.trim())
// v0.12.0: pass whatever token the widget on the passcode
// step has produced. If the operator has Turnstile required
// and the user hasn't solved the passcode-step widget, the
// backend refuses and we bounce them back to email-entry
// with a clear status (see catch below).
await requestOtc(email.trim(), { turnstileToken })
setTurnstileToken(null)
setStep('code')
setStatus(
'Too many failed attempts. We sent a one-time code to your email — use it to sign in.',
)
} catch (e2) {
setTurnstileToken(null)
if (e2.status === 429) {
setStep('code')
setStatus(
'Too many failed attempts. Wait a minute, then request a one-time code to sign in.',
)
} else if (e2.status === 400) {
setStep('email')
setStatus(
'Too many failed attempts. Solve the challenge below to receive a one-time code.',
)
} else {
setStatus(
'Too many failed attempts. Use the "Use a code instead" link to sign in via email.',
@@ -146,24 +263,74 @@ export default function Login() {
setBusy(true)
setStatus('')
try {
await verifyOtc(email.trim(), code.trim())
// OTC verified. If the user has no passcode, offer to set one
// before redirecting. We re-read `has_passcode` from the server
// rather than caching the step-1 result because the user could
// have set a passcode in another tab between then and now.
const { has_passcode } = await checkPasscode(email.trim())
if (has_passcode) {
window.location.assign('/')
} else {
setStep('offer-passcode')
setBusy(false)
await verifyOtc(email.trim(), code.trim(), { trustDevice })
// OTC verified the server has signed in the user. Fetch the
// canonical /api/auth/me to decide where to land:
// * needs_profile §6.1 capture (then /beta-pending).
// * no passcode §6.2 offer-passcode (then /).
// * otherwise /.
// needs_profile wins over has_passcode: a pending user can't yet
// do anything that benefits from faster sign-in, so we don't
// distract them with the passcode offer mid-admission.
const meResp = await fetch('/api/auth/me', { credentials: 'include' })
let me = null
if (meResp.ok) {
try {
me = await meResp.json()
} catch (_) {
me = null
}
}
if (me?.needs_profile === true) {
setStep('capture-profile')
setStatus('')
setBusy(false)
return
}
if (me?.has_passcode === false) {
setStep('offer-passcode')
setStatus('')
setBusy(false)
return
}
// Either /me returned the granted-with-passcode shape, or the
// call failed but the session cookie is set fall through to
// a hard reload so App.jsx re-fetches and renders accordingly.
window.location.assign('/')
} catch (err) {
setStatus('That code is invalid or expired. Try again, or request a new code.')
setBusy(false)
}
}
async function submitProfile(e) {
if (e) e.preventDefault()
const fn = firstName.trim()
const ln = lastName.trim()
const why = reason.trim()
if (!fn || !ln || !why) {
setStatus('All three fields are required.')
return
}
setBusy(true)
setStatus('')
try {
await submitBetaRequest({
first_name: fn,
last_name: ln,
beta_request_reason: why,
})
// Hard-load so App.jsx re-fetches /api/auth/me and picks up
// the captured fields. The user stays permission_state='pending'
// until an admin grants access the next thing they should
// see is the "your request is in review" page.
window.location.assign('/beta-pending')
} catch (err) {
setStatus(err.message || 'Could not submit your request. Try again.')
setBusy(false)
}
}
async function submitNewPasscode(e) {
if (e) e.preventDefault()
const pc = newPasscode.trim()
@@ -197,6 +364,21 @@ export default function Login() {
}
}
function onReasonKey(e) {
// §6.1 ergonomic: Cmd/Ctrl+Enter submits the capture form from
// the reason textarea (the multi-line input that would otherwise
// swallow Enter as a newline).
if ((e.metaKey || e.ctrlKey) && e.key === 'Enter') {
submitProfile(e)
}
}
function onNewPasscodeKey(e) {
if ((e.metaKey || e.ctrlKey) && e.key === 'Enter') {
submitNewPasscode(e)
}
}
function backToEmail() {
setStep('email')
setCode('')
@@ -206,17 +388,27 @@ export default function Login() {
async function fallbackToOtc() {
// Manual "Use a code instead" from the passcode step. Same shape
// as the email-step OTC dispatch.
// as the email-step OTC dispatch. v0.12.0: pass through whatever
// Turnstile token the passcode-step widget has produced (or null
// when the widget is disabled at build time).
setBusy(true)
setStatus('')
try {
await requestOtc(email.trim())
await requestOtc(email.trim(), { turnstileToken })
setTurnstileToken(null)
setPasscode('')
setStep('code')
setStatus('Check your inbox — a six-digit code is on the way.')
} catch (err) {
setTurnstileToken(null)
if (err.status === 429) {
setStatus('Slow down — wait a minute before requesting another code.')
} else if (err.status === 400) {
// v0.12.0: the widget rejected or no token was sent. Bounce
// the user back to the email step so they get a fresh
// challenge alongside the email input.
setStep('email')
setStatus("Couldn't verify you're human. Please retry the challenge.")
} else {
setStatus(err.message || 'Could not request a code. Try again.')
}
@@ -249,7 +441,18 @@ export default function Login() {
required
disabled={busy}
/>
<button type="submit" disabled={busy || !email.trim()}>
{/*
v0.12.0: CloudFlare Turnstile widget. Renders nothing
when VITE_TURNSTILE_SITE_KEY is unset (and turnstileReady
defaults to true in that case so the submit gate doesn't
lock up). On every challenge the widget calls onToken
with the fresh token; we feed it to /auth/otc/request.
*/}
<TurnstileWidget onToken={setTurnstileToken} />
<button
type="submit"
disabled={busy || !email.trim() || !turnstileReady}
>
{busy ? 'Checking…' : 'Continue'}
</button>
</form>
@@ -270,6 +473,30 @@ export default function Login() {
required
disabled={busy}
/>
{/* v0.11.0 trust device for 30 days. The checkbox lives
on the verify step so the user makes the trust gesture
in the same breath as signing in. Off by default; the
user opts in deliberately. */}
<label className="otc-trust-device">
<input
type="checkbox"
checked={trustDevice}
onChange={e => setTrustDevice(e.target.checked)}
disabled={busy}
/>
<span>Trust this device for 30 days</span>
</label>
{/*
v0.12.0: a second Turnstile widget for the
"Use a code instead" fallback dispatch. The passcode
verify path does not consume a Turnstile token (it has
its own 5-attempt lockout shape from v0.10.0), but if
the user falls back to OTC the same /auth/otc/request
endpoint runs and needs a token. We render the widget
on this step too so the fallback works without bouncing
back to email-entry first.
*/}
<TurnstileWidget onToken={setTurnstileToken} />
<div className="otc-actions">
<button type="submit" disabled={busy || !passcode.trim()}>
{busy ? 'Signing in…' : 'Sign in'}
@@ -278,7 +505,7 @@ export default function Login() {
type="button"
className="btn-link-quiet"
onClick={fallbackToOtc}
disabled={busy}
disabled={busy || !turnstileReady}
>
Use a code instead
</button>
@@ -312,6 +539,17 @@ export default function Login() {
required
disabled={busy}
/>
{/* v0.11.0 trust device for 30 days. Same shape as the
passcode step; the user opts in deliberately. */}
<label className="otc-trust-device">
<input
type="checkbox"
checked={trustDevice}
onChange={e => setTrustDevice(e.target.checked)}
disabled={busy}
/>
<span>Trust this device for 30 days</span>
</label>
<div className="otc-actions">
<button type="submit" disabled={busy || code.length !== 6}>
{busy ? 'Signing in…' : 'Sign in'}
@@ -330,6 +568,60 @@ export default function Login() {
</p>
</form>
)}
{step === 'capture-profile' && (
<form onSubmit={submitProfile}>
<p className="otc-hint">
You're signed in. {import.meta.env.VITE_APP_NAME} is in private
beta tell us a bit about yourself and an admin will review
your request.
</p>
<label className="otc-field-label">First name</label>
<input
ref={firstNameRef}
type="text"
autoComplete="given-name"
value={firstName}
onChange={e => setFirstName(e.target.value)}
required
disabled={busy}
maxLength={120}
/>
<label className="otc-field-label">Last name</label>
<input
type="text"
autoComplete="family-name"
value={lastName}
onChange={e => setLastName(e.target.value)}
required
disabled={busy}
maxLength={120}
/>
<label className="otc-field-label">
Why you'd like to be included in the beta
</label>
<textarea
value={reason}
onChange={e => setReason(e.target.value)}
onKeyDown={onReasonKey}
required
disabled={busy}
rows={5}
maxLength={4000}
placeholder="A sentence or two is plenty."
/>
<div className="otc-actions">
<button
type="submit"
disabled={busy || !firstName.trim() || !lastName.trim() || !reason.trim()}
>
{busy ? 'Submitting…' : 'Submit request'}
</button>
</div>
<p className="otc-shortcut-hint">
Tip: <kbd></kbd>+<kbd>Enter</kbd> (or <kbd>Ctrl</kbd>+<kbd>Enter</kbd>) to submit.
</p>
</form>
)}
{step === 'offer-passcode' && (
<div className="otc-offer-passcode">
<p className="otc-hint">
@@ -366,6 +658,7 @@ export default function Login() {
autoComplete="new-password"
value={newPasscode}
onChange={e => setNewPasscode(e.target.value)}
onKeyDown={onNewPasscodeKey}
placeholder="New passcode"
required
disabled={busy}
@@ -388,11 +681,13 @@ export default function Login() {
</form>
)}
{status && <p className="otc-status">{status}</p>}
<p className="otc-fallback">
<Link to="/philosophy">Read the philosophy </Link>
<span className="otc-fallback-sep">·</span>
<a href="/auth/login">Sign in with Gitea (fallback)</a>
</p>
{step !== 'capture-profile' && (
<p className="otc-fallback">
<Link to="/philosophy">Read the philosophy </Link>
<span className="otc-fallback-sep">·</span>
<a href="/auth/login">Sign in with Gitea (fallback)</a>
</p>
)}
</div>
</div>
)
@@ -32,6 +32,9 @@ import {
getMe,
setPasscode,
clearPasscode,
listMyDevices,
revokeMyDevice,
revokeAllMyDevices,
} from '../api.js'
import { getConsent, onConsentChange, hydrateFromServer } from '../lib/consent.js'
@@ -54,11 +57,126 @@ export default function NotificationSettings({ viewer }) {
<WatchesSection />
<MutesSection viewer={viewer} />
<SignInSection />
<DevicesSection />
<PrivacyCookiesSection />
</div>
)
}
// v0.11.0: trusted devices (§6.2, roadmap item #9)
//
// Lists the user's active device-trust rows and lets them revoke any
// or all. A revoke marks the row dead server-side; the matching
// device's next visit will be refused and the cookie cleared. The
// surface intentionally does not single out the row whose cookie the
// current request carries every row reads identically, so the user
// can revoke "this device" alongside any other from a single page.
function DevicesSection() {
const [devices, setDevices] = useState(null)
const [error, setError] = useState(null)
const [busy, setBusy] = useState(false)
useEffect(() => {
refresh()
}, [])
async function refresh() {
try {
const { items } = await listMyDevices()
setDevices(items || [])
setError(null)
} catch (e) {
setError(e.message || 'Could not load trusted devices.')
}
}
async function onRevoke(deviceId) {
setBusy(true)
try {
await revokeMyDevice(deviceId)
await refresh()
} catch (e) {
setError(e.message || 'Could not revoke device.')
} finally {
setBusy(false)
}
}
async function onRevokeAll() {
if (!confirm('Revoke trust on every device, including this one? You will be asked to sign in via email next time.')) {
return
}
setBusy(true)
try {
await revokeAllMyDevices()
await refresh()
} catch (e) {
setError(e.message || 'Could not revoke devices.')
} finally {
setBusy(false)
}
}
return (
<SectionShell
title="Trusted devices"
subtitle="Devices where you've checked “Trust this device for 30 days.” Sign-in is automatic on these devices until the trust expires or you revoke it."
>
{devices === null && <p className="settings-note">Loading</p>}
{devices !== null && devices.length === 0 && (
<p className="device-empty">
No trusted devices. Sign in and check Trust this device for 30 days
to add the device you're on now.
</p>
)}
{devices !== null && devices.length > 0 && (
<>
<ul className="device-list">
{devices.map(d => (
<li key={d.id} className="device-list-item">
<div className="device-meta">
<div className="device-ua">{d.user_agent || 'Unknown device'}</div>
<div className="device-stamps">
Trusted {formatStamp(d.created_at)} · last seen {formatStamp(d.last_seen_at)} · expires {formatStamp(d.expires_at)}
</div>
</div>
<button
type="button"
onClick={() => onRevoke(d.id)}
disabled={busy}
title="Revoke trust on this device"
>
Revoke
</button>
</li>
))}
</ul>
<button
type="button"
className="device-revoke-all"
onClick={onRevokeAll}
disabled={busy}
>
Revoke all devices
</button>
</>
)}
{error && <p className="settings-note warning">{error}</p>}
</SectionShell>
)
}
function formatStamp(stamp) {
// The server emits SQLite `datetime('now')` strings (UTC, no
// timezone marker). Parse defensively; fall back to the raw stamp
// if Date can't make sense of it.
if (!stamp) return '—'
const d = new Date(stamp.replace(' ', 'T') + 'Z')
if (Number.isNaN(d.getTime())) return stamp
return d.toLocaleString()
}
// §6.2 sign-in (v0.10.0 / roadmap item #8): passcode management
function SignInSection() {
+120
View File
@@ -0,0 +1,120 @@
// TurnstileWidget.jsx v0.12.0 / roadmap item #10.
//
// Renders the CloudFlare Turnstile JS widget on the email-entry step of
// `/login`. Reads the site key from `import.meta.env.VITE_TURNSTILE_SITE_KEY`
// (Vite convention VITE_* prefix is build-time embedded). When the
// site key is unset/empty, this component renders nothing and reports
// a `null` token through `onToken` so the parent form can still submit.
// The backend's `TURNSTILE_REQUIRED` policy decides what happens to a
// request that arrives without a token; the frontend is intentionally
// not in that loop. See `backend/app/turnstile.py` for the matrix.
//
// The CloudFlare script is loaded once per page on first widget mount.
// Subsequent mounts (e.g. user goes back to email-entry after a failed
// OTC request) reuse the script tag and re-render the widget on the
// fresh container `div`. Unmounting removes the widget instance via
// `turnstile.remove(widgetId)` so a remount produces a new challenge
// rather than reusing a stale, already-consumed token.
//
// Turnstile contract:
// * `data-callback` fires with the token string on a successful
// challenge; the token is single-use and expires after ~5 minutes.
// * `data-error-callback` fires on a failed challenge (network,
// blocked, etc.); we surface a `null` token so the parent shows
// a retry hint.
// * `data-expired-callback` fires when the token times out before
// submission; we also drop to `null` and re-render so the user
// gets a fresh challenge on retry.
//
// We do **not** import the CloudFlare script at build time; loading it
// dynamically here keeps the bundle clean of an external request the
// page may not need (anonymous viewers reading RFCs never see Login).
import { useEffect, useRef } from 'react'
const TURNSTILE_SCRIPT_URL = 'https://challenges.cloudflare.com/turnstile/v0/api.js'
const SITE_KEY = import.meta.env.VITE_TURNSTILE_SITE_KEY || ''
// Promise-keyed: only one script tag, only one resolution chain.
let scriptLoadPromise = null
function loadTurnstileScript() {
if (typeof window === 'undefined') return Promise.resolve(null)
if (window.turnstile) return Promise.resolve(window.turnstile)
if (scriptLoadPromise) return scriptLoadPromise
scriptLoadPromise = new Promise((resolve, reject) => {
const existing = document.querySelector(`script[src="${TURNSTILE_SCRIPT_URL}"]`)
if (existing) {
existing.addEventListener('load', () => resolve(window.turnstile))
existing.addEventListener('error', reject)
return
}
const script = document.createElement('script')
script.src = TURNSTILE_SCRIPT_URL
script.async = true
script.defer = true
script.addEventListener('load', () => resolve(window.turnstile))
script.addEventListener('error', reject)
document.head.appendChild(script)
})
return scriptLoadPromise
}
export function turnstileEnabled() {
return !!SITE_KEY
}
export default function TurnstileWidget({ onToken, theme = 'auto' }) {
const containerRef = useRef(null)
const widgetIdRef = useRef(null)
useEffect(() => {
if (!SITE_KEY) {
// No site key configured surface a null token immediately so
// the parent form's submit-disabled gate doesn't lock up
// waiting on a challenge that will never arrive. The backend
// decides whether a tokenless request is admitted.
onToken?.(null)
return undefined
}
let cancelled = false
loadTurnstileScript()
.then(turnstile => {
if (cancelled || !turnstile || !containerRef.current) return
widgetIdRef.current = turnstile.render(containerRef.current, {
sitekey: SITE_KEY,
theme,
callback: token => onToken?.(token),
'error-callback': () => onToken?.(null),
'expired-callback': () => onToken?.(null),
})
})
.catch(() => {
// Script load failure surface null so the parent can decide
// what to do (today: still let submit through; the backend
// policy decides admission).
if (!cancelled) onToken?.(null)
})
return () => {
cancelled = true
if (widgetIdRef.current && window.turnstile) {
try {
window.turnstile.remove(widgetIdRef.current)
} catch (_) {
// Already gone or never registered nothing to clean up.
}
widgetIdRef.current = null
}
}
// We intentionally do not list `onToken` in the dependency array;
// a parent re-rendering with a fresh closure should not tear down
// and rebuild the widget (which would consume a fresh challenge).
// eslint-disable-next-line react-hooks/exhaustive-deps
}, [])
if (!SITE_KEY) return null
return <div ref={containerRef} className="turnstile-widget" />
}